Android snapshot helper: a capture cancelled in its first milliseconds stalls the next capture #4030

Found while working on #4021 (#4005). This happens on main's sources and was not introduced by that PR.

Symptom. In CI Coverage, packages/platform-android/src/__tests__/snapshot-helper-session.test.ts › "cancels a stalled snapshot, retires its helper, and starts the next capture cleanly" timed out at 5 s (run 40711232553).

Trigger. The test aborts the stalled capture with a fixed 10 ms timer. When the same flow is aborted 0–2 ms in instead, the next capture hangs for more than 3 s. This reproduces on main's sources and on #4021 alike. On a slow CI host the 10 ms abort lands early in the flow, which turns this into an intermittent timeout.

Possible product impact. If a real snapshot is cancelled right after it starts, the next capture on that device may stall instead of starting cleanly. That fits the "persistent session didn't answer" runs reported in #4005 and #3929.

Required behavior.

• A capture cancelled at any point leaves the helper retired or reusable.
• The next capture starts within its normal budget.
• The test drives the abort at the risky points explicitly, rather than through a wall-clock timer.