fix: run src/browser unit tests in CI and stabilize the vendor digest

[Agnik47]

Description

The unit vitest project excluded src/browser/**/*.test.ts, and no other project picked those files up — plugin covers plugins/*/test/**, e2e lists individual files under tests/e2e/, and smoke covers tests/smoke/**. The result was that 44 test files / 596 tests covering CDP, snapshots, the Playwright sandbox and the local Cloak runtime ran in no CI job at all.

Removing the exclude also restores vitest's default excludes (node_modules, dist) for that project, and matches the other four projects, none of which set exclude.

One small correction to the issue: the exclude came from #196, not #216. #216 only dropped the clis/** half of the list during the adapter migration:

commit	unit exclude
c08cb276 (init)	['clis/**/*.test.{ts,js}']
b51cac37 (#196)	['clis/**/*.test.{ts,js}', 'src/browser/**/*.test.ts']
0470ac43 (#216)	['src/browser/**/*.test.ts']

Why the two extra changes are here

Enabling the tests surfaces a pre-existing failure in src/browser/run/playwright-client-build.test.ts, which shells out to build-playwright-sandbox-client.mjs --check. Without fixing it, this PR would turn the windows-latest unit job red. Two independent causes:

1. The vendor digest was platform-dependent. vendorDigest() hashes each vendored file's relative path, built with path.join(), so entries hash as client\artifact.ts on Windows and client/artifact.ts everywhere else. The check could therefore never pass on Windows:

as-is (native separator) : 882f0880d4838423647cc617ff199298536b606803fd8dbcdb3cb5a8bed31131
POSIX-normalized         : 23a97afd31faa0e5a64b116e897c3b90123eafcc2ca9102586a062eaf533350d
vendor-manifest.json     : 23a97afd31faa0e5a64b116e897c3b90123eafcc2ca9102586a062eaf533350d

Normalizing the separator to / is a no-op on POSIX, so the pinned hash is unchanged and vendor-manifest.json needs no regeneration.

2. The vendored sources and generated bundle are compared byte-for-byte. On a checkout with core.autocrlf=true (the Git for Windows default, and there was no .gitattributes), CRLF breaks both the digest and the separate "bundle is up to date" comparison — the latter with a misleading Generated Playwright QuickJS client is out of date error. .gitattributes pins just those two paths to LF.

Both fixes are necessary; I verified each one independently fails without the other.

Related issue: #231

Type of Change

• 🐛 Bug fix
• ✨ New feature
• 🌐 New site adapter
• 📝 Documentation
• ♻️ Refactor
• 🔧 CI / build / tooling

Checklist

• I ran the checks relevant to this PR
• I updated tests or docs if needed
• I included output or screenshots when useful

Screenshots / Output

npm run typecheck, npm run build and the git diff --exit-code generated-artifact check all pass.

All 44 previously-skipped files now run and pass through the real unit project (Windows, Node 22):

$ npx vitest run --project unit src/browser/
 Test Files  44 passed (44)
      Tests  594 passed | 2 skipped (596)

Full unit project, before and after — same pre-existing failures, no new ones:

before:  Test Files  4 failed | 91 passed (95)     Tests  15 failed | 1544 passed | 5 skipped (1564)
after:   Test Files  4 failed | 135 passed (139)   Tests  15 failed | 2138 passed | 7 skipped (2160)

The 4 failing files (hosted/runner, hosted/files, plugin, docs-sync-review-cli) fail identically on unmodified main on this machine and pass in CI; they're a local environment artifact (my temp dir lives under $HOME, so paths render as ~\AppData\...). The two plugin-project failures (pypi, instagram) also reproduce on clean main.

One thing I could not verify

I don't have Bun locally, so I could not run the bun-test job, which also uses --project unit and will now execute these 44 files. QuickJSHost (WASM via quickjs-emscripten) and the test that spawns process.execPath look like the likeliest friction points under Bun. Flagging it rather than assuming — happy to adjust if that job goes red.

Notes

• No new test added deliberately: the guard for the digest regression is playwright-client-build.test.ts itself, which now actually runs on Windows in CI. Adding a dedicated test would mean exporting vendorDigest() out of the provenance script purely for testability — happy to do it if you'd prefer.
• .gitattributes is intentionally scoped to the two byte-sensitive paths rather than a repo-wide * text=auto.
• Existing Windows clones will see those vendored files re-checkout as LF once this lands.

[github-actions]

🟢 No documentation gap found — medium confidence

The automated review found no documentation gap in the supplied changes.

This review is advisory and does not block merging.