fix: recover Android recordings after daemon restart

[thymikee]

Summary

Fixes #1117 by recovering Android screenrecord when daemon recording state is lost before record stop.

Adds device-side recovery metadata plus a bounded process-scan fallback so a fresh daemon can stop, pull, and clean the orphaned recording instead of returning no active recording.

Validation

• pnpm format
• pnpm typecheck
• pnpm build
• pnpm vitest run --project unit src/daemon/handlers/tests/record-trace.test.ts
• pnpm vitest run --project provider-integration test/integration/provider-scenarios/android-recording.test.ts
• pnpm test:integration:provider
• pnpm test:smoke
• Emulator verification on Pixel 9 Pro XL API 37 / emulator-5554: start recording, scroll Settings, wipe daemon state mid-recording, then record stop from a fresh daemon recovered and produced a non-empty MP4.

Attempted pnpm check:unit and pnpm test:coverage; both hit unrelated host-load/platform flakes, and the failing files/groups passed when rerun in isolation.

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.5 MB	1.5 MB	+6.5 kB
JS gzip	492.6 kB	494.1 kB	+1.4 kB
npm tarball	588.6 kB	590.0 kB	+1.5 kB
npm unpacked	2.1 MB	2.1 MB	+6.5 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	26.8 ms	27.5 ms	+0.7 ms
CLI --help	50.7 ms	50.6 ms	-0.1 ms

Top changed chunks:

Chunk	Raw diff	Gzip diff
dist/src/record-trace-recording.js	+6.5 kB	+1.4 kB

[thymikee]

Ready for human review.

Review pass on #1129:

• GitHub CI is clean: 21/21 checks passing.
• The original Fallow failure was fixed by bed08021c.
• The recovery path is intentionally conservative: it only runs when daemon recording state is missing and no session still has active recording state, validates metadata against a live Android screenrecord process, uses bounded ADB probes, and refuses ambiguous recovery when multiple agent-device recordings are active.
• The PR includes provider integration coverage plus live emulator evidence: Pixel 9 Pro XL API 37 / emulator-5554, recording recovered after daemon state was wiped mid-recording and produced a non-empty MP4.

Residual risk: this is the narrow recovery fix for #1117. Durable chunk/session ownership hardening is intentionally stacked in #1131 / #1130, so it should follow this PR rather than block it.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-07-06 13:28 UTC