test(e2e): Run live delivery specs in parallel

[utkarshcloudinary]

No issue. CI speed-up, follow-up to #1330.

Approach

The E2E job is the longest CI job, and its single-worker @serial pass (18 tests, about 1.8 min) is the biggest part. Five of the eight @serial files were tagged only because they need the real Cloudinary connection, which every analytics spec overwrites with fake credentials. They do not break each other: they write the same real URL, and each one changes only settings that the others do not check. This PR moves them to a new multi-worker pass.

• New tag @live on the image, video, gallery, SVG and asset-edit delivery specs.
• npm run test:e2e now runs three passes: analytics and other safe specs (3 workers), @live (3 workers), then @serial (1 worker).
• The 3 files that really change the whole site stay @serial: deactivation (it uninstalls the plugin), wizard setup, and connection analytics.
• New tests/e2e/utils/live-site.js. It makes the steps that act on site-wide state take turns through a file lock:
  • wp cloudinary sync, because it is not scoped to one attachment. Its analyze step clears and re-adds _cld_unsynced on every attachment, so two parallel syncs can make one of them find nothing.
  • The gallery's /cloudinary/v1/image_data sync.
  • Post cleanup.
  • Settings writes to the shared media_display option (SVG settings and video_player).
• The lock is an atomic mkdir in artifacts/. It records the owner PID, so a worker that dies while holding it does not block the run.

I tested the lock alone with 4 processes and 320 lock sections (sync and async callbacks): no overlaps. A lock from a dead PID was taken over at once.

QA notes

• In the E2E job log, npm run test:e2e should show three Playwright runs: about 20 tests on 3 workers, 9 @live tests on 3 workers, and 9 @serial tests on 1 worker.
• Compare the "Run E2E tests" step time with recent develop runs (about 2.6 to 2.8 min).
• To run only the new pass locally: npm run test:e2e:live.

[utkarshcloudinary]

Closing. The CI run showed no gain: the E2E step took 2:53 against 2:33 to 2:48 on develop. The site lock makes the slow sync steps run one at a time, so the @LiVe pass saved only about 6s. It also made asset-edit flaky, because the SVG spec changes image_format/image_quality while asset-edit compares full delivery URLs. A better next step is separate WordPress sites per pass (the second wp-env site, or 2 CI runners).