fix(db): dual-SQLite WAL corruption when nativeDb reads after better-sqlite3 writes

[carlos-alm]

Bug

The WASM engine path crashes during buildStructure with:

Error: queryGet row failed: database disk image is malformed

Root cause

setupPipeline always opens both ctx.db (better-sqlite3, bundled SQLite) and ctx.nativeDb (rusqlite, separate bundled SQLite) on the same DB file, regardless of engine mode.

When engineName === 'wasm':

1. insertNodes writes via better-sqlite3 (because it gates on ctx.engineName === 'native')
2. buildStructure reads via ctx.nativeDb.queryGet() (no engine gate — just checks if nativeDb exists)

Two different bundled SQLite copies accessing the same WAL file causes corruption. The rusqlite connection reads WAL state written by better-sqlite3's SQLite copy, and the WAL checksums/formats may be subtly incompatible.

Impact

• WASM engine builds fail with SQLite corruption when native addon is also available
• v3.5.0 build benchmark: both engines failed (WASM from this bug, native from refactor: remove dead WASM call-AST extraction and pre-3.2 edge shim #686)

Fix

Gate nativeDb usage in read stages (buildStructure, finalize, etc.) on ctx.engineName === 'native' to ensure consistent SQLite library usage within a single build. When engine is 'wasm', all DB operations should go through better-sqlite3.