feat: exec engine, Kokoro TTS, OCR tag, model architecture

- Run All notebook engine — 11 runtime adapters, progress bar, abort, SQLite context store
- Kokoro TTS hybrid engine — English/Chinese ONNX + Web Speech API fallback for 10+ languages
- OCR tag ({{@ocr:}}) — image-to-text with Text/Math/Table modes, amber card styling
- Architecture-aware ai-worker — qwen3 vs qwen3_5 model loading, auto-fallback to HuggingFace
- Vision support flags on Qwen 3.5 Flash, 35B-A3B, DeepSeek V3.2
- Language Learning template with TTS pronunciation tips
- SQLite-compatible SQL in Technical template
- model-hosts.js for configurable ONNX model hosting
- scripts/mirror-models.sh for self-hosted model mirroring
- Deleted moonshine-medium-worker.js (replaced by speech-worker.js)
- 12 new Playwright tests (191 total)

Author: ijbo

CHANGELOG-exec-engine.md

@@ -0,0 +1,147 @@
+# Exec Engine, Kokoro TTS, OCR, Model Hosts & Templates
+
+- Added **"Run All" button** to the formatting toolbar — executes every code/tag block in document order
+- Built **Block Registry** (`exec-registry.js`) with FNV-1a content-hash stable IDs and central adapter discovery
+- Built **Execution Controller** (`exec-controller.js`) with progress bar, abort support, per-block status badges (pending/running/done/error), and event emitter
+- Added **SQLite-backed shared context store** (`_exec_results` table) — blocks can write results, downstream blocks read them via `$(blockId)` references or SQL queries
+- Registered **11 runtime adapters**: `bash`, `math`, `python`, `html`, `javascript`, `sql`, `docgen-ai`, `docgen-image`, `docgen-agent`, `api`, `linux-script`
+- DocGen adapters (AI/Image/Agent) use **auto-accept mode** during Run All — skip review panel, replace tag with generated content directly
+- API adapter **auto-accepts** fetch results — replaces `{{API:}}` tag with response markdown
+- Linux adapter submits code to **Judge0 CE** API directly and returns stdout/stderr
+- Added **deferred adapter queue** (`M._pendingRuntimeAdapters`) to handle module loading order
+- Added `css/exec-engine.css` — Run All button (green gradient), fixed-bottom progress bar, per-block status badges
+- Added **Kokoro TTS hybrid engine** (`textToSpeech.js`) — English/Chinese via Kokoro 82M ONNX, Japanese & others via Web Speech API fallback
+- Added **TTS WebWorker** (`tts-worker.js`) — runs `kokoro-js` off main thread for jank-free synthesis; voice auto-selection by language
+- Added `css/tts.css` — TTS button styling, loading states, speaker icon animations
+- Added **model-hosts.js** — configurable model hosting with `MODEL_HOST_FALLBACK` (HuggingFace) for automatic fallback when primary mirror fails
+- Updated **ai-worker.js** — architecture-aware model loading (`qwen3` vs `qwen3_5`), per-component dtype config, `setModelId` accepts `architecture` and `dtype` params, automatic fallback to HuggingFace on primary host failure
+- Added **Kokoro TTS** entry to `ai-models.js` — `kokoro-tts` with `isTtsModel: true` flag, ~80 MB download
+- Added `supportsVision: true` to Qwen 3.5 Flash, Qwen 3.5 35B-A3B, and DeepSeek V3.2 cloud models
+- Added **OCR tag type** to DocGen — `{{@OCR:}}` tag with amber-accented card, Qwen model default, OCR mode pills (Text/Math/Table), image upload with `@upload:` sync
+- Added **OCR card CSS** (`ai-docgen.css`) — amber color scheme, mode pill selector, dark/light theme support
+- Added `data-ocr-mode`, `data-mode`, `data-upload-index` to DOMPurify allowlist
+- Deleted `js/moonshine-medium-worker.js` — replaced by unified `speech-worker.js`
+- Updated `speech-worker.js` — accepts `MODEL_ARCH` and `MODEL_DTYPE` configuration for flexible model loading
+- Added **Language Learning template** (`agents.js`) — multilingual lesson generator with vocabulary, dialogue, grammar, quiz, TTS pronunciation tips
+- Fixed: SQLite-compatible SQL in **Technical template** — replaced `SERIAL`/`VARCHAR`/`NOW()` with `INTEGER`/`TEXT`/`datetime('now')`
+- Updated **Documentation template** — added Run All notebook section, updated test count to 191
+- Added `scripts/mirror-models.sh` — shell script for self-hosting ONNX models on GitLab
+- Updated `package.json` — added `kokoro-js` dependency
+- Added `public/assets/demos/25_run_all.png` — Run All demo screenshot
+- Added **Run All** entry to Help Mode (`help-mode.js`)
+- Added **Run All** entry to Feature Demos (`feature-demos.js`)
+- Added **12 Playwright tests** for the execution engine (`tests/feature/exec-engine.spec.js`)
+- All 191 existing tests pass with zero regressions
+
+---
+
+## Summary
+
+This release adds three major features: (1) a unified "Run All" notebook engine that executes every code block, AI tag, API call, and Linux compile in document order with visual progress tracking; (2) a hybrid Kokoro TTS engine for high-quality text-to-speech in English, Chinese, and 10+ languages via Web Speech API fallback; and (3) an OCR tag type for image-to-text extraction. Supporting changes include configurable model hosting with automatic fallback, architecture-aware AI worker loading, vision support flags on cloud models, and an updated template library.
+
+---
+
+## 1. Block Registry
+**Files:** `js/exec-registry.js` [NEW]
+**What:** Central registry that scans the document for all executable blocks (fenced code blocks + DocGen/API/Linux tags). Assigns stable content-hash-based IDs using FNV-1a hashing. Maintains ordered block list and maps blocks to their runtime adapters. Drains a pending adapter queue (`M._pendingRuntimeAdapters`) on init.
+**Impact:** Every block type is discoverable and addressable by a stable ID, enabling the controller to orchestrate execution.
+
+## 2. Execution Controller
+**Files:** `js/exec-controller.js` [NEW]
+**What:** Orchestrates "Run All" — scans via registry, executes sequentially, resolves `$(blockId)` cross-references, manages progress bar (fixed bottom), per-block status badges, and abort flow. Emits lifecycle events.
+**Impact:** One-click execution of entire document with visual progress tracking and abort capability.
+
+## 3. SQLite Context Store
+**Files:** `js/exec-sandbox.js`
+**What:** Auto-creates `_exec_results` table in the existing sql.js instance. Each block's output is stored as key-value pair. SQL blocks can natively query previous results. Added HTML, JS, and SQL runtime adapters.
+**Impact:** Blocks share data within a single Run All session.
+
+## 4. DocGen Runtime Adapters (Auto-Accept)
+**Files:** `js/ai-docgen-generate.js`
+**What:** Registered `docgen-ai`, `docgen-image`, and `docgen-agent` adapters. During Run All, AI/Image/Agent blocks auto-replace the tag with generated content directly.
+**Impact:** AI/Image/Agent tags execute during Run All without manual review.
+
+## 5. API & Linux Runtime Adapters
+**Files:** `js/api-docgen.js`, `js/linux-docgen.js`
+**What:** API adapter parses config, executes `fetch()`, auto-replaces `{{API:}}` tag. Linux adapter submits to Judge0 CE.
+**Impact:** API calls and compiled code participate in Run All.
+
+## 6. Code Runtime Adapters
+**Files:** `js/executable-blocks.js`, `js/exec-math.js`, `js/exec-python.js`
+**What:** Registered `bash`, `math`, and `python` adapters wrapping existing execution logic.
+**Impact:** All 6 in-browser code runtimes participate in Run All.
+
+## 7. Kokoro TTS Hybrid Engine
+**Files:** `js/textToSpeech.js` [NEW], `js/tts-worker.js` [NEW], `css/tts.css` [NEW]
+**What:** Hybrid text-to-speech using Kokoro 82M v1.1-zh ONNX for English/Chinese (high quality, off-thread via WebWorker) and Web Speech API for Japanese and all other languages. Audio playback via AudioContext with proper cleanup. Voice auto-selection by language code.
+**Impact:** Pronunciation preview on any text in preview — hover and click 🔊 to hear it spoken.
+
+## 8. Model Hosting & AI Worker Architecture
+**Files:** `js/model-hosts.js` [NEW], `ai-worker.js`, `public/ai-worker.js`
+**What:** Configurable model hosting with `MODEL_HOST_FALLBACK` to HuggingFace. AI worker now accepts `architecture` (qwen3 vs qwen3_5) and `dtype` params for flexible model loading. Automatic fallback on primary host failure.
+**Impact:** Models can be self-hosted with transparent fallback; supports both text-only and vision model architectures.
+
+## 9. OCR Tag Type
+**Files:** `js/ai-docgen.js`, `css/ai-docgen.css`, `js/renderer.js`
+**What:** New `{{@OCR:}}` tag with amber-accented card, OCR mode pills (Text/Math/Table), image upload with `@upload:` sync back to editor. Qwen model default. DOMPurify allowlist expanded.
+**Impact:** Extract text from images directly in markdown documents.
+
+## 10. Vision Support & Model Registry
+**Files:** `js/ai-models.js`
+**What:** Added `supportsVision: true` to Qwen 3.5 Flash, 35B-A3B, and DeepSeek V3.2. Added `kokoro-tts` model entry with `isTtsModel: true` flag.
+**Impact:** Vision-capable models shown in image/OCR card selectors; TTS model discoverable in model registry.
+
+## 11. Templates & Documentation
+**Files:** `js/templates/agents.js`, `js/templates/ai.js`, `js/templates/documentation.js`, `js/templates/technical.js`
+**What:** New Language Learning template with vocabulary, dialogue, grammar, quiz, and TTS pronunciation tips. SQLite-compatible SQL in Technical template. Documentation template updated with Run All section and test count.
+**Impact:** Better template coverage; SQL templates now actually run in the SQLite sandbox.
+
+## 12. Infrastructure
+**Files:** `scripts/mirror-models.sh` [NEW], `js/speech-worker.js`, `index.html`, `src/main.js`, `package.json`
+**What:** Model mirroring shell script for self-hosting ONNX models. Speech worker accepts architecture/dtype config. New modules loaded in Phase 3b. `kokoro-js` dependency added. Moonshine medium worker deleted.
+**Impact:** Self-hosting option for air-gapped deployments; cleaner speech module architecture.
+
+---
+
+## Files Changed (38 total)
+
+| File | Lines Changed | Type |
+|------|:---:|------|
+| `js/exec-registry.js` | +239 | New: Block scanner, adapter registry, FNV-1a IDs |
+| `js/exec-controller.js` | +383 | New: Run All orchestration, progress UI, abort |
+| `js/textToSpeech.js` | +251 | New: Hybrid TTS engine (Kokoro + Web Speech) |
+| `js/tts-worker.js` | +136 | New: Kokoro TTS WebWorker |
+| `js/model-hosts.js` | +19 | New: Configurable model hosting with fallback |
+| `css/exec-engine.css` | +146 | New: Run All button, progress bar, status badges |
+| `css/tts.css` | +178 | New: TTS button styling, animations |
+| `tests/feature/exec-engine.spec.js` | +161 | New: 12 Playwright tests |
+| `scripts/mirror-models.sh` | +130 | New: ONNX model mirroring script |
+| `public/assets/demos/25_run_all.png` | — | New: Run All demo screenshot |
+| `js/ai-docgen-generate.js` | +280 | DocGen auto-accept adapters |
+| `js/ai-docgen.js` | +377 −5 | OCR tag type, block scanner exports |
+| `js/exec-sandbox.js` | +179 | SQLite context store, HTML/JS/SQL adapters |
+| `ai-worker.js` | +126 −86 | Architecture-aware model loading, fallback |
+| `public/ai-worker.js` | +126 −86 | Mirror of ai-worker.js |
+| `css/ai-docgen.css` | +114 | OCR card styling, mode pills |
+| `js/templates/agents.js` | +160 −118 | Language Learning template |
+| `js/templates/ai.js` | +121 | TTS pronunciation tips in templates |
+| `js/api-docgen.js` | +77 | API runtime adapter |
+| `js/linux-docgen.js` | +63 | Linux runtime adapter |
+| `js/speech-worker.js` | +61 −36 | Architecture/dtype config |
+| `js/ai-models.js` | +39 | Vision flags, Kokoro TTS entry |
+| `js/exec-math.js` | +33 | Math adapter |
+| `js/exec-python.js` | +32 | Python adapter |
+| `js/executable-blocks.js` | +22 | Bash adapter |
+| `js/templates/documentation.js` | +20 −3 | Run All docs, test count update |
+| `js/model-hosts.js` | +19 | Model hosting config |
+| `package-lock.json` | +17 | kokoro-js dependency |
+| `js/templates/technical.js` | +16 −10 | SQLite-compatible SQL |
+| `index.html` | +11 | Run All button, TTS module |
+| `src/main.js` | +7 | Module loading |
+| `js/help-mode.js` | +6 | Run All help entry |
+| `js/renderer.js` | +3 −1 | DOMPurify attr expansion |
+| `js/ai-assistant.js` | +2 −2 | Minor fix |
+| `js/feature-demos.js` | +1 | Run All demo entry |
+| `package.json` | +1 | kokoro-js dependency |
+| `README.md` | +5 −3 | Release notes (pre-existing) |
+| `js/moonshine-medium-worker.js` | −393 | Deleted: replaced by speech-worker |