fix: remove deadlock in pt_usleep sleep-spam blocker (can hang render thread → os_unfair_lock crash)

[ap-harsa-yamani]

Summary

Fixes a deadlock in the pt_usleep sleep-spam blocker (blockSleepSpamming toggle) that can hang a thread forever — and, when the hung thread is a render/GPU thread, corrupts os_unfair_lock and crashes the game.

The bug

In PlayTools/PlayLoader.m, pt_usleep:

static int pt_usleep(useconds_t time) {
    dispatch_once(&thread_sleep_once, ^{
        ...
        thread_sleep_lock = [[NSLock alloc] init];
        [thread_sleep_lock lock];      // (1) LOCKED inside dispatch_once, never unlocked
    });

    if ([[PlaySettings shared] blockSleepSpamming]) {
        ...
        if (exceeded_sleep_limit) {     // >100× usleep(100000) within 2s
            [thread_sleep_lock lock];   // (2) NSLock is NOT recursive → DEADLOCK forever
            [thread_sleep_lock unlock]; //     never reached
            return 0;                   //     never reached
        }
    }
    return usleep(time);
}

• NSLock is not recursive: the lock taken inside dispatch_once (1) is held forever, so the second lock (2) blocks the thread permanently instead of "stopping" it.
• If that thread is UnityGfxDeviceWorker / a render thread / the main thread, the render pipeline hangs → os_unfair_lock corruption → crash.
• This matches the crash signature reported for Arknights: Endfield (BUG IN CLIENT OF LIBPLATFORM: Trying to recursively lock an os_unfair_lock, EXC_BREAKPOINT/SIGKILL), see [Game Support]: Arknights: Endfield GPU-related thread lock corruption causes crash PlayCover#2067 and [Bug]: Arknights Endfield 1.4.3 silent-exits after ~60s on M3/8GB macOS 26.6.1 — no crash report, all bypasses exhausted PlayCover#2202.

The fix

• Remove the deadlocking thread_sleep_lock entirely. The per-thread counter is already protected by @synchronized(thread_sleep_counters).
• When a thread exceeds the usleep limit, skip its sleeps (return 0) instead of hanging it — same anti-spam effect, no deadlock.
• Remove the now-unneeded UIApplicationWillTerminateNotification observer that unlocked the (now removed) lock.

Testing

• Brace/paren balance verified; the change is purely deletion of the lock + its observer.
• Could not run a full xcodebuild here (no iOS SDK toolchain available in this environment); the diff is minimal and self-contained.
• Recommended verification: build PlayTools, install, enable blockSleepSpamming in a Unity game (e.g. Endfield), and confirm the game no longer hangs/crashes after ~60s.

Related

• [Game Support]: Arknights: Endfield GPU-related thread lock corruption causes crash PlayCover#2067 — "Arknights: Endfield GPU-related thread lock corruption causes crash"
• [Bug]: Arknights Endfield 1.4.3 silent-exits after ~60s on M3/8GB macOS 26.6.1 — no crash report, all bypasses exhausted PlayCover#2202 — "Arknights Endfield 1.4.3 silent-exits after ~60s" (fresh M3 data)

[TheMoonThatRises]

@ohaiibuzzle not sure what the original design intention was for the lock system was