feat(test): replace 'deferred' retry strategy with 'isolated'

[dgozman]

Summary

• retryStrategy: 'isolated' runs retries at the end of the run, one by one in a single worker, replacing the unreleased 'deferred' value.
• Tests that did not have a chance to run because of a failure keep their priority in the queue instead of being deferred together with retries.

[github-actions]

Test results for "MCP"

1 failed
❌ [firefox] › mcp/annotate.spec.ts:447 › should switch screencast to -s session on show --annotate @mcp-macos-latest-firefox

7759 passed, 1249 skipped

---

Merge workflow run.

[github-actions]

Test results for "tests 1"

7 flaky

⚠️ [chromium-library] › library/browsercontext-page-event.spec.ts:173 › should work with Ctrl-clicking `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/chromium/chromium.spec.ts:177 › serviceWorker(), and fromServiceWorker() work `@realtime-time-library-chromium-linux`
⚠️ [chromium-library] › library/video.spec.ts:356 › screencast › should work for popups `@realtime-time-library-chromium-linux`
⚠️ [chromium-library] › library/video.spec.ts:736 › screencast › should work with video+trace `@realtime-time-library-chromium-linux`
⚠️ [chromium-page] › page/workers.spec.ts:191 › should attribute network activity for worker inside iframe to the iframe `@chromium-ubuntu-22.04-node24`
⚠️ [playwright-test] › ui-mode-test-watch.spec.ts:185 › should watch new file `@ubuntu-latest-node26`
⚠️ [playwright-test] › ui-mode-trace.spec.ts:388 › should reveal errors in the sourcetab `@windows-latest-node22`

49698 passed, 1156 skipped

---

Merge workflow run.

[github-actions]

Hi, I'm the Playwright bot and I took a first look at the CI failures here.

🟢 The one failure is a pre-existing flake — this PR is clear

The only failing test is [firefox] › mcp/annotate.spec.ts:447 › should switch screencast to -s session on show --annotate, and it's a known flake that fails across main and unrelated PRs. This PR only touches the test-runner dispatcher and retry strategy, which doesn't exercise the MCP screencast/annotate path.

Details

Overall: 1 failed in the MCP workflow, 7759 passed. No failure is caused by this PR.

Pre-existing flake / infra

• [firefox] › mcp/annotate.spec.ts:447 › should switch screencast to -s session on show --annotate (@mcp-macos-latest-firefox) — pre-existing flake. In the aggregated test-results DB this exact (file, test_title) fails on many SHAs this PR can't be responsible for: main pushes (e.g. runs 28955218357, 29128339906, 29199182758) and unrelated PRs (fix(cli): don't crash the response when navigating closes the last tab #41675, feat(devops): add PR CI failure triage bot #41616, test(bidi): install libavcodec in CI for Firefox #41736, chore(protocol): make timeout a protocol-level field #41712, fix(video): avoid MJPEG interlace inference #41776, fix(ci): keep Copilot GitHub access read-only #41791). It flips verdict on every browser — firefox 4/196 runs, chromium 2/200, chrome 2/201, webkit 1/202. On the Firefox MCP bot alone it's 13 failed of 191 runs (~7%), with this PR's run being the latest red square:

  🟩🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥🟥

  The diff (dispatcher.ts retry queueing, config.ts/configLoader.ts renaming deferred→isolated, retry.spec.ts) never reaches the MCP screencast flow, so there's no plausible path from this change to that failure.

_{Triaged by the Playwright bot - agent run}