chore(trace): remove snapshot pointers from action events

[Dmitry Gozman (dgozman)]

Summary

• Actions no longer point at their DOM snapshots. Snapshots are addressed by callId + phase, the same way screenshots and aria snapshots already are.
• Removes BeforeActionTraceEvent.pageId, beforeSnapshot, inputSnapshot and afterSnapshot, plus the now unused CallMetadata.pageId / CallMetadata.frameId.
• Page entries in the model come from page open/close events; internal pages are no longer reported to the instrumentation.
• trace snapshot takes --phase <before|action|after> instead of --name <before|input|after>.

[github-actions]

Test results for "tests 1"

1 failed
❌ [chromium-library] › library/video.spec.ts:664 › screencast › should capture full viewport @chromium-ubuntu-22.04-node22

3 flaky

⚠️ [chromium-library] › library/video.spec.ts:664 › screencast › should capture full viewport `@chromium-ubuntu-22.04-arm-node20`
⚠️ [chromium-library] › library/popup.spec.ts:260 › should not throw when click closes popup `@chromium-ubuntu-22.04-node24`
⚠️ [webkit-library] › library/browsercontext-storage-state.spec.ts:543 › should not leave IndexedDB connections open `@webkit-ubuntu-22.04-node20`

51249 passed, 1240 skipped

---

Merge workflow run.

[github-actions]

Test results for "MCP"

3 failed
❌ [chromium] › mcp/clipboard.spec.ts:19 › clipboard write without permission dialog @mcp-macos-latest-chromium
❌ [chromium] › mcp/clipboard.spec.ts:19 › clipboard write without permission dialog @mcp-windows-latest-chromium
❌ [chromium] › mcp/clipboard.spec.ts:19 › clipboard write without permission dialog @mcp-ubuntu-latest-chromium

8256 passed, 1361 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 screencast flake — this PR looks clear

library/video.spec.ts:664 › screencast › should capture full viewport failed on chromium-ubuntu-22.04-node22, but it's a well-known pixel-timing flake and the identical failure rescued itself on retry in this very same run on another bot. This PR removes snapshot pointers from action events (trace addressing by callId + phase) — it doesn't touch video/screencast capture. The 3 flaky results need no triage.

Details

This PR changes the trace model, snapshotter, trace-viewer and the trace snapshot CLI. The failing test asserts on captured screencast frame pixels, which none of that code path exercises.

Pre-existing flake / infra

• [chromium-library] › library/video.spec.ts:664 › screencast › should capture full viewport@chromium-ubuntu-22.04-node22 — the test fails with Expected all pixels to satisfy isAlmostRed, found bad pixel (255, 255, 255, 255): a frame that should be solid red is momentarily white, the classic screencast render-timing flake. Across the test-results DB this test is chronically rescued on retry — 334 retry-rescues and only 1 final failure ever recorded (this run). It flakes on push (main-branch, no PR) and unrelated PRs alike, so the PR can't be responsible.

  The proof it's the same flake here: in this same run (33158602402) it failed on chromium-arm-node20 too with the identical isAlmostRed signature, but a retry rescued it there (so it's reported only as flaky) — on node22 it happened to exhaust all four attempts and went red. Same flake, worse luck.

  Final-verdict history on chromium-ubuntu-22.04-node22 (this PR is the trailing 🟥; the next main run right after is green):

  🟩🟩🟥🟩

  History on chromium-ubuntu-22.04-arm-node20 (🟧 = a retry rescued a failed attempt — 65 of 653 runs; this PR is the trailing 🟧):

  🟧🟩

Not triaged (flaky, passed on retry)

• [chromium-library] › library/video.spec.ts:664 › screencast › should capture full viewport@chromium-ubuntu-22.04-arm-node20, [chromium-library] › library/popup.spec.ts:260 › should not throw when click closes popup@chromium-ubuntu-22.04-node24, [webkit-library] › library/browsercontext-storage-state.spec.ts:543 › should not leave IndexedDB connections open@webkit-ubuntu-22.04-node20 — reported flaky, all rescued on retry, none in the trace/snapshot code paths this PR changes.

_{Triaged by the Playwright bot - agent run}

[github-actions]

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

🟢 Both failures are pre-existing flakes/infra — this PR looks clear

The latest runs report a video.spec.ts screencast failure and three mcp/clipboard.spec.ts failures. Both fail the same way on main and on unrelated PRs, so neither is caused by this PR. This PR only reworks trace snapshot addressing (action events, snapshotter, trace-viewer, trace snapshot CLI) — it doesn't touch video recording or the MCP clipboard flow.

Details

Pre-existing flake / infra

• [chromium-library] › library/video.spec.ts:664 › screencast › should capture full viewport@chromium-ubuntu-22.04-node22 — a classic screencast pixel-timing flake: the frame comes back white (bad pixel (255,255,255,255), expected almost-red). It flaked (rescued on retry) on @chromium-ubuntu-22.04-arm-node20 in this very run, and across the results DB it's rescued on retry 299 times on shas unrelated to this PR, failing across many bots (arm, macos, ubuntu node20/22/24). 2639 passing final verdicts vs this single first-attempt failure. Nothing this PR changes affects screencast pixels.

• [chromium] › mcp/clipboard.spec.ts:19 › clipboard write without permission dialog@mcp-{ubuntu,windows,macos}-latest-chromium — same error on all three OSes: the clipboard reads back "" instead of "Hello from Playwright!". It's currently failing everywhere, not just here — same failure on the main push 6748802c (this PR's own base, no PR involved) in run 33157511907, and on unrelated PRs #42440, #42441, #42442 and #42434 within the same window. It's an environment/permission issue (first seen 2026-08-21, spiking now), not something this PR could cause.

Failing on main and four other PRs at the same time is the tell — this is shared infra noise, not a regression from this PR.

_{Triaged by the Playwright bot - agent run}