parking_lot.c should use _PyTime_GetMonotonicClock() when available

[colesbury]

Bug report

The _PySemaphore_PlatformWait function in parking_lot.c has three implementations:

• WaitForSingleObjectEx (for Windows)
• sem_timedwait (Linux and most other POSIX platforms)
• pthread_cond_timedwait (macOS and possibly other platforms that don't have sem_timedwait)

We should follow the pattern used in thread_pthread.h where we prefer functions that support CLOCK_MONOTONIC. These are sem_clockwait and pthread_cond_timedwait after setting pthread_condattr_setclock.

Additionally, we should use _PyTime_AsTimespec_clamp instead of _PyTime_AsTimespec (we don't want to raise errors in these code paths).

Note that:

1. Windows doesn't need any changes because WaitForSingleObjectEx isn't relative to any specific clock -- it just takes a timeout for N milliseconds in the future
2. macOS doesn't support pthread_condattr_setclock, so we're stuck with the system clock on macOS. It's not clear to me if we use the pthread_cond_timedwait for any other OS.

Linked PRs

• gh-112606: Use monotonic clock when supported in semaphore waiting in parking_lot.c #112614
• gh-112606: Use pthread_cond_timedwait_relative_np in parking_lot.c when available #112616
• gh-112606: Use sem_clockwait with monotonic time when supported in parking_lot.c #112733

[mattprodani]

Splitting into two PRs for sem_timedwait and pthread_cond_timedwait implementations. Working on finishing up the condvar implementation to share fields that initialize the proper pthread_condattr_t from thread_pthread.h as per notes.

• gh-112606: Use sem_clockwait with monotonic time when supported in parking_lot.c #112733
• gh-112606: Use pthread_cond_timedwait_relative_np in parking_lot.c when available #112616

[vstinner]

As someone writes a fix for the mutex, ping me, I can review it ;-)

[colesbury]

Closing this as the relevant PRs landed a while ago.