feat: unique test results per run (--run-name)

[stanleykim-ux]

Stacked on #2 (feat/parallel-local-devices). Review only the last commit.

Why

Two appwright runs on one machine (for example yarn test:ios and yarn test:android side by side, which #2 makes practical) clobber each other's output: Playwright wipes test-results/ at the start of every run, the html reporter deletes playwright-report/ before rebuilding it, and appwright's own video / worker-info store was hardcoded to playwright-report/data/videos-store (there was a TODO in workerInfo.ts asking for exactly this fix).

What

Every run writes into its own folders:

test-results/<run>/               Playwright output (per-test artifacts, .last-run.json)
test-results/<run>/videos-store/  appwright worker videos and worker-info files
playwright-report/<run>/          HTML report  ->  npx playwright show-report playwright-report/<run>

• appwright test --run-name <name> names the run. The flag is stripped by the appwright CLI (Playwright rejects unknown options) and passed down as APPWRIGHT_RUN_NAME, which can also be set directly (CI).
• Default name: <project>-<YYYYMMDD>-<HHmmss>-<4 random chars>, e.g. android-20260910-153915-k3i4. Sortable and tells you which platform at a glance.
• Names are sanitized to [A-Za-z0-9._+-] so they are always a safe path segment.
• A custom outputDir / html outputFolder in the consumer config is kept as the base; the run name is nested under it.
• Global setup logs one line with the run's folders so interleaved output from concurrent runs can be told apart.
• Consumer (tulip-player-mobile/e2e) needs no change: yarn test:android --run-name smoke passes straight through.

Notes for reviewers

• Playwright evaluates the config file again inside every worker process, so defineConfig cannot generate a fresh name; resolveRunName() reads APPWRIGHT_RUN_NAME and only generates (and stores) one when it is unset. That also makes the direct npx playwright test -c appwright.config.ts path work.
• The video store moves from playwright-report/data/videos-store to test-results/<run>/videos-store. The html reporter removes its whole output folder in onEnd before it reads attachment files, so anything stored under playwright-report/ was lost at report time. Playwright clears outputDir at run start, so the new location is fresh per run.
• --last-failed reads .last-run.json from the run's folder, so it needs the same --run-name again. Documented.
• Run folders are never deleted automatically. Retention is deliberately out of scope; documented as rm -rf test-results playwright-report.

Verification

• npm run build, npm test (75 tests, 38 new covering name generation/sanitizing, flag parsing, reporter mapping, CLI invocation).
• Real runs against a booted Android emulator with the example Wikipedia APK via a scratch project pointing at the built dist/:
  • appwright test --project android --run-name smoke → test-results/smoke/{.last-run.json,videos-store/worker-info-0.json}, playwright-report/smoke/index.html.
  • appwright test --project android (no flag) → android-20260910-153915-k3i4.
  • npx playwright test --config appwright.config.ts --project android (bypassing the appwright CLI) → android-20260910-153922-r2em.
  • All three folder sets survived side by side.
• Not run: two runs literally concurrent on two platforms (only one emulator handy). Per-run isolation follows from the folder layout above, so the risk is low.

🤖 Generated with Claude Code