fix: add cold-restart native-driver regression check to electron smoke (#7592)

[diegosouzapw]

Refs #7592

Root cause (verified at current tip)

The original theory in #7592 (better-sqlite3 shipped inside app.asar) was
refuted — better-sqlite3 ships via extraResources, never inside app.asar.
The real mechanism the reporters converged on — a Turbopack-hashed
better-sqlite3-<hash> import resolving a stale Node-ABI binary instead of
the canonical Electron-ABI one — is already fixed on this tip:

• PR fix(electron): bump electron 42→43 + build better-sqlite3 from source (ABI 148) #6605 (merged into main): Electron 42→43, rebuild better-sqlite3 from
  source at ABI 148.
• PR fix(electron): normalize hashed standalone externals #7353 (merged into release/v3.8.49): normalize Turbopack-hashed
  standalone externals so better-sqlite3-<hash> resolves to the canonical
  package.

Both merge commits are confirmed ancestors of the current tip via
git merge-base --is-ancestor. The sql.js OOM/retry-loop compounding issue is
also capped (#6835/#6628/#6632).

The issue stayed open only because the regression test was missing: a
community reporter's own investigation on the issue thread independently
reached the same conclusion and explicitly flagged that
scripts/dev/smoke-electron-packaged.mjs never exercises a packaged app's
second launch against an existing DB, and never asserts which SQLite
driver loaded.

What this PR does (test-only, no production code change)

• scripts/dev/smoke-electron-packaged.mjs: extract the launch/wait/settle
  logic into launchAndCollectLogs() and add an
  ELECTRON_SMOKE_COLD_RESTART=1 mode that relaunches the packaged app
  against the same, persisted DATA_DIR and asserts (new exported
  assertNativeDriverSelected()) that the startup log shows a native driver
  (bun:sqlite / better-sqlite3 / node:sqlite) selected, not the sql.js
  (WASM) fallback.
• .github/workflows/electron-release.yml: wire ELECTRON_SMOKE_COLD_RESTART=1
  into the Linux smoke leg (blocking — Linux has no Windows-specific sandbox
  caveats), while leaving the Windows leg as-is per the existing
  continue-on-error comment.
• tests/unit/electron-smoke-script.test.ts: unit tests for
  assertNativeDriverSelected() covering all three native drivers, the
  sql.js-fallback failure case, and the missing-driver-line failure case.

Validation

This is analyzed/implemented per the triage plan-file's needs-vps verdict —
a full electron-builder packaged run (native rebuild + code-signing, ~GB
output) is out of scope for this sandbox, so the new orchestration logic
(launchAndCollectLogs, cold-restart wiring) is not exercised end-to-end
here. What is verified locally:

• node --import tsx/esm --test tests/unit/electron-smoke-script.test.ts —
  7/7 passing, including the 3 new assertNativeDriverSelected cases (native
  driver accepted, sql.js fallback flagged, missing driver line flagged).
• node --check scripts/dev/smoke-electron-packaged.mjs — syntax OK.
• npm run typecheck:core — clean.
• npx eslint --suppressions-location config/quality/eslint-suppressions.json <changed files> — clean.
• node scripts/check/check-file-size.mjs — OK (script isn't in the scanned dirs; no regression).
• node scripts/check/check-complexity.mjs / check-cognitive-complexity.mjs — both OK, ratchet improved (no new violations).
• node scripts/check/check-changelog-integrity.mjs — OK.

The actual two-launch-with-driver-assertion path will get its first real
exercise on the Linux leg of electron-release.yml (xvfb-run, packaged
build) the next time that workflow runs — recommend tracking CI results there
before closing #7592 outright.

Citations

• fix(startup): better-sqlite3 not unpacked from app.asar → native module load fails → sql.js fallback OOM crash #7592
• fix(electron): bump electron 42→43 + build better-sqlite3 from source (ABI 148) #6605
• fix(electron): normalize hashed standalone externals #7353
• _tasks/pipeline/bugs/2-implementing/7592-better-sqlite3-asar-unpack-sqljs-oom.plan.md