fix(ios): let a preset play again after it has been stopped

[piaskowyk]

What

The Audio Sync demo in PulsarApp plays once. When the preset reaches its end and you hit Play again, nothing happens — no sound, no haptics.

playFrom() in the demo stops before every play. On the first play that stop is a no-op (nothing has parsed yet). On a replay at the same position it is not: PresetHandle.ensureParsed reuses the parsed composer, so the same CHHapticPatternPlayers are reused — and stopPlayer(id:) kept them in the registry after stopping them, so playPlayer called start(atTime: 0) on players CoreHaptics had already stopped. That silently does nothing, and since the discrete player carries the preset's audio event, both the haptics and the audio go quiet.

Seeking kept working because a new position re-parses and builds fresh players. Same-position replay is the only broken path, which is also why it slipped past qa-demos-audio-sync.yaml — that flow taps replay, but it can only assert the JS clock, not whether anything actually played.

Changes

• HapticEngineWrapper — stopping a player unregisters it, so the next playPlayer takes the existing rebuild-from-pattern path (which also restarts the engine if it went down). The registered audio resource is untouched, so the rebuilt player still carries its audio.
• PatternComposer — parse() releases the players it replaces. They were leaked into the engine registry on every re-parse, so roughly ten seeks were enough for eviction to start stopping players still in use.

Verified on hardware

Confirmed on a physical iPhone 13 (iOS 26.6.1) by A/B-ing two builds of PulsarApp that differ by exactly one line — the unregisterPlayer(id) call in stopPlayer(id:) — with the demo's JS untouched in both:

build	Play, let it finish, Play again
with the call	audio + haptics play
without it	silent

Worth stating plainly: a stopped CHHapticPatternPlayer does not start again. That is the behaviour this change works around, and it is now measured rather than assumed.

Tests

• New playingAfterAStopBuildsAFreshPlayer in the CoreHaptics mock suite. It fails on main (playersCreated → 1) == 2) and passes here.
• Pulsar-Package: 63 tests / 11 suites pass, plus the 7-test isolated mock suite.

[piaskowyk]

Closing pending a proper investigation. The A/B held up on hardware, but it was a fix arrived at by inference and then confirmed, not one derived from understanding the failure — and that is not a good enough basis for changing engine-wide player lifecycle. Reopening or replacing this once the actual mechanism is established with a debugger.