chore(recorder): render highlights via the shared element highlight

[Dmitry Gozman (dgozman)]

Summary

• Recorder highlights the current selector through the shared element highlight (Frame.addHighlight) instead of the UIState.
• New internal aria-template=<json> selector engine, so that the highlighted aria template goes through the same code path.
• Trace viewer highlights via injectedScript.addHighlight() directly.
• Shared element highlight fixes: the action point element is created lazily, the rAF loop ignores glass panes and tolerates unqueryable selectors, and internal calls are not tracked as actions.
• Highlights are confined to the frame the selector resolves to for now; cross-frame highlighting will be restored together with piercing frames support.

[github-actions]

Test results for "tests 1"

3 flaky

⚠️ [chromium-library] › library/chromium/chromium.spec.ts:213 › should intercept service worker requests (main and within) `@chromium-ubuntu-22.04-node22`
⚠️ [firefox-page] › page/page-emulate-media.spec.ts:144 › should keep reduced motion and color emulation after reload `@firefox-ubuntu-22.04-node20`
⚠️ [firefox-page] › page/page-goto.spec.ts:90 › should work with Cross-Origin-Opener-Policy `@firefox-ubuntu-22.04-node20`

51122 passed, 1208 skipped

---

Merge workflow run.

[github-actions]

Test results for "MCP"

3 failed
❌ [firefox] › mcp/cli-core.spec.ts:77 › click with --modifiers @mcp-windows-latest-firefox
❌ [firefox] › mcp/cli-devtools.spec.ts:84 › request shows full request and response details @mcp-windows-latest-firefox
❌ [firefox] › mcp/cli-devtools.spec.ts:173 › response-body returns just the body @mcp-windows-latest-firefox

8098 passed, 1311 skipped

---

Merge workflow run.

[github-actions]

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

🟢 CI is clear — the 3 failures are pre-existing Firefox flakes

The three failing MCP tests all flip verdict on unrelated PRs and on main, and this PR only touches recorder highlighting and selector/frame code that the MCP CLI flow doesn't exercise. Nothing here looks caused by the change.

Details

The tests 1 report has 0 failures (3 flaky, retried green) — nothing to triage. The MCP report has 3 failures, all [firefox] on mcp-windows-latest-firefox.

Pre-existing flake / infra

• [firefox] › mcp/cli-core.spec.ts:77 › click with --modifiers — Firefox-only flake: 3 failed / 1981 runs (99.8% pass). Also failed on unrelated PRs fix(mcp): validate Host and Origin on CDP relay WebSocket upgrades #42103 (3d7e075e) and test(filechooser): allow exact one second lastModified rounding #42145 (5bc1c77d). Passes on chromium/webkit/chrome/msedge every time.
• [firefox] › mcp/cli-devtools.spec.ts:84 › request shows full request and response details — Firefox-only flake: 7 failed / 1981 runs. Failed on a main push (4b0cc996) and PRs Revert "feat(routeFromHar): add interceptAPIRequests option (#41294)" #41964, feat(chromium): roll to r1236 #42038, test(filechooser): allow exact one second lastModified rounding #42145, chore: remove experimental component testing #42168 — none related to this change.
• [firefox] › mcp/cli-devtools.spec.ts:173 › response-body returns just the body — Firefox-only flake: 6 failed / 1981 runs. Failed on a main push (1e6a5e63) and PRs fix(mcp): launch the Chrome profile that has the extension installed #41939, fix(cli): rewrite CLI command in installed skills #42137.

All three are green on every other browser (chromium, webkit, chrome, msedge: 0 failures across ~2000 runs each), and this PR's diff — shared element highlight, aria-template= selector engine, recorder/trace-viewer highlight plumbing — doesn't reach the MCP CLI click/request/response paths.

_{Triaged by the Playwright bot - agent run}