test: stabilize iOS snapshot timeout regression

[thymikee]

Summary

Stabilize the iOS snapshot timeout lifecycle regression tests.

The injected modal probe previously self-released after eight seconds while recovery through live SpringBoard accessibility could take roughly fifteen seconds on loaded CI hosts. The test then inspected state after the abandoned work had already drained and falsely failed its busy and retention assertions.

The test now uses a deterministic unavailable application target and keeps the injected probe blocked until it records the abandoned-work state. It then explicitly releases the probe and verifies the real drain path. Both snapshotFast and snapshotRaw still run through their production entrypoints.

Validation

• Reproduced the original assertions locally by shortening the old self-release window.
• Ran both fixed tests for 10 iterations each: 20 tests passed.
• pnpm format
• AGENT_DEVICE_XCUITEST_INCLUDE_UNIT_TESTS=1 pnpm build:xcuitest
• pnpm check:affected --base origin/main --run
• git diff --check

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.8 MB	1.8 MB	-2.7 kB
JS gzip	567.2 kB	566.2 kB	-939 B
npm tarball	681.8 kB	680.5 kB	-1.3 kB
npm unpacked	2.4 MB	2.4 MB	-3.7 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	26.0 ms	25.9 ms	-0.1 ms
CLI --help	53.2 ms	53.6 ms	+0.4 ms

Top changed chunks:

Chunk	Raw diff	Gzip diff
dist/src/registry.js	-1.1 kB	-301 B
dist/src/cli-help.js	-655 B	-218 B
dist/src/session.js	-140 B	-88 B
dist/src/internal/daemon.js	-159 B	-86 B
dist/src/snapshot-diagnostics.js	-254 B	-83 B

[thymikee]

Exact-head review (f785c5b3): clean. The change is confined to the unit-test gate and does not affect shipping behavior. Both real snapshotFast and snapshotRaw entrypoints still exercise the production bounded-probe, abandoned-work, penalty, recovery-before-drain, and onDrained paths. Replacing live SpringBoard with an unavailable deterministic target prevents fallback-tier duration from racing the old self-release, while the test-owned semaphore and defer preserve bounded cleanup and revert sensitivity. All checks are green, including iOS Smoke and Swift unit compile. Applying ready-for-human despite draft status under coordinator policy.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-07-16 19:00 UTC