perf(e2e): Run specs in parallel workers and overlap CI setup

[utkarshcloudinary]

Improves e2e test runtime in CI (no linked issue).

Approach

• The e2e job took 5.5 to 6.5 min: ~2.5 min for the Playwright run with a single worker, ~2 min for setup run strictly in sequence. This PR makes the analytics specs worker-safe so most of the suite runs in parallel, and overlaps the independent setup steps.
• Per-worker analytics capture log. The mu-plugin reads a cld_e2e_worker cookie (browser and REST traffic) or a CLD_E2E_WORKER env var (WP-CLI via docker exec) and appends to analytics-capture-<worker>.log. Unmarked requests keep using the shared log. A cookie rather than extraHTTPHeaders because the header is attached cross-origin and forced CORS preflights that third parties (fonts inside the Cloudinary player iframe) rejected.
• New tests/e2e/fixtures.js sets both markers per worker; all specs import test/expect from it.
• workers: 3. Specs that mutate site-wide state (wizard-setup, connection-analytics, deactivation-analytics, image and video delivery) are tagged @serial. npm run test:e2e runs the parallel set, then the serial set with one worker, in the same wp-env.
• settings-analytics flips image_quality instead of image_format so it no longer races media-analytics on the same key.
• Admin API calls made with the fake e2e cloud are answered locally with a 401. Connect::history() issued seven ~1s real round-trips per dashboard load, which pushed page=cloudinary past the navigation timeout.
• cache-analytics: create the cache point after the admin page load, and clear the stale parent plus the asset lock before the single-purge test. activate_parents() deletes the unconfigured parent and leaves a 10s _asset_lock transient that blocks the tracked branch in rest_purge_all().
• CI, e2e job: wp-env start runs in the background while browsers install and assets build; Playwright browsers cached by version; Setup PHP and a Composer cache added before npm ci. Without Setup PHP the e2e job's npm ci took 2.5 to 4 min on the same runs where every job with it took ~35s.
• CI, unit job: install only @wordpress/env (391 packages, ~15s) at the lockfile version into a scratch prefix instead of the full npm ci; Composer deps installed directly. The wp-env sources cache is removed: once it actually restored, it carried the previous run's install state, the plugin activation hook did not run, and PHPUnit flagged two tests risky for a missing relationships table.
• Tried and reverted: npm ci --ignore-scripts made the install step take 4 to 7 min with npm 10.

CI timings, latest run on this branch vs median of the 5 runs before it:

step	before	after
E2E job total	363s (327 to 391)	302s
npm ci	40s	39s
setup (browsers, build, wp-env)	127s	104s
Run E2E tests	147s	119s (parallel 50s + serial 66s)
Unit job total	187s	121s

Adding new specs:

• Parallel (default). Import test/expect from ./fixtures; nothing else is needed. Use fakeCloudinaryConnected(), assert only on findAnalyticsEvents( events, 'name' ) rather than the total event count, and own the settings keys the spec flips (check rg "image_settings\[" tests/e2e before picking one).

• Serial. Tag the describe block and say why in a comment; npm run test:e2e runs tagged specs in a second, single-worker pass after the parallel set:

  // @serial: writes real credentials to `cloudinary_connect`.
  test.describe( 'Real sync flow', { tag: '@serial' }, () => {

  Tag it if the spec writes or deletes cloudinary_connect (any ensureCloudinaryConnected() or wp cloudinary sync), deactivates or uninstalls the plugin, or flips a global setting another spec reads without restoring it. Serial specs may still run alongside nothing else, so they are free to assume the plugin is active and connected as they left it.

• Tests within one file always run in order in both modes (fullyParallel is off), so per-file beforeAll/afterAll state is safe.

• Run one phase with npm run test:e2e:parallel or npm run test:e2e:serial; one file with npx playwright test --config tests/e2e/playwright.config.js my.spec.js.

QA notes

• Compare the E2E job timings on this PR against a recent develop run (gh run view <id> --json jobs).
• Locally: npm run env:start && npm run build && npm run test:e2e. Both phases should pass. docker exec <tests-cli> ls wp-content/uploads/ should show analytics-capture-w0.log through w2.log.
• Run npm run test:e2e twice in a row to confirm the cache-analytics ordering fix holds across runs.
• Open the video delivery test with --headed or check the Playwright trace: no CORS errors for fonts.gstatic.com in the console.