feat(ci): Vivado in GitHub Actions Docker — Closes #604

[gHashTag]

Summary

Replaces the Railway self-hosted runner approach (PR #621) with a pure GitHub Actions Docker setup. No self-hosted runner, no Railway costs, no external infra.

Architecture

build-vivado-image.yml  (one-time, manual)
  -> fetches Vivado_2025.2_Lin64.bin from a private GH Release asset
  -> silent-installs Vivado ML Standard, Artix-7 family only (~35 GB)
  -> pushes ghcr.io/<owner>/t27-vivado:2025.2

vivado-synth.yml  (per PR + workflow_dispatch)
  container: ghcr.io/<owner>/t27-vivado:2025.2
  -> builds t27c
  -> t27c fpga-build --smoke    (Verilog from .t27 specs)
  -> vivado -mode batch          (synth + P&R + bitstream)
  -> uploads .bit + utilization summary as artifacts

Files

• infra/vivado-docker/Dockerfile — multi-stage Ubuntu 22.04 + Vivado 2025.2 install. Strips all non-Artix-7 device families and discards installer stage.
• infra/vivado-docker/install_config.txt — silent-install config (Modules=Vivado:1, Artix-7:1, others:0, DocNav:0, EnableDiskUsageOptimization=1).
• infra/vivado-docker/README.md — operator runbook (2-step setup: release upload + workflow click) and troubleshooting.
• .github/workflows/build-vivado-image.yml — manual image-build workflow with easimon/maximize-build-space to free ~45 GB before install. Pushes to GHCR with GHA layer cache.
• .github/workflows/vivado-synth.yml — synth pipeline using container: directive. Triggers on PRs touching fpga/, specs/fpga/, bootstrap/src/, and on manual dispatch with design selection (blinky / gf16 / phi_heartbeat) and UART toggle.
• Removed obsolete .github/workflows/vivado-bitstream.yml (used dead kkrizka/vivado:2019.2 image).

Operator steps after merge

1. Upload Vivado installer once as a private GH Release asset:

   gh release create vivado-installer-2025.2 \
     --repo gHashTag/t27 \
     --prerelease \
     ~/Downloads/Vivado_2025.2_Lin64.bin
   

2. Actions tab → 'Build Vivado Docker Image' → Run workflow (one-time, ~60-90 min).
3. Subsequent PRs to fpga/** auto-trigger vivado-synth.yml and produce bitstream artifacts.

Why this beats Railway

	Railway runner (#621)	GHA Docker (this PR)
Monthly cost	$35–55	$0
Setup steps	9 manual (CLI + UI)	2 (release upload + click)
Vivado upgrade	SSH into container	Re-run workflow
Cold start	Always-on	3–5 min image pull

Hardware-in-the-loop tok/s still requires DLC-10 JTAG attached to a real host. That stays the trios-bridge → Tailscale path on the Mac (gaia-macbook-air.tail2c3a29.ts.net): synth in GHA → bridge endpoint flashes the FPGA → uart-smoke binary streams telemetry back.

Policy compliance

• no .sh / .py (only Dockerfile + YAML with inline awk)
• no Cyrillic in docs/architecture/specs paths
• Refs feat(fpga): Vivado CI pipeline for gf16_heartbeat_uart_top (UART telemetry on K20) #604 (Vivado-CI blocker), Refs infra: Railway Vivado self-hosted GH Actions runner #620 (Railway issue, now superseded)

Supersedes

• PR feat(infra): Railway Vivado self-hosted GH Actions runner #621 (Railway runner — to close)
• Issue infra: Railway Vivado self-hosted GH Actions runner #620 (Railway provisioning — to close)

Refs #604
Refs #620

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.

[github-actions]

📓 NotebookLM Notebook linked to this PR

• Notebook ID: b83263109fb055dc
• URL: https://notebooklm.google.com/notebook/b83263109fb055dc

This notebook contains session context, decisions, and artifacts for this work.