After fork(), only the calling thread survives. Locks held by
dead parent threads stay locked forever, causing deadlocks when
the child tries to acquire them.

- Add reinit_pymutex_after_fork / reinit_pyrwlock_after_fork
  utilities that zero the raw lock bytes directly
- Reset all PyGlobalState locks, thread-handle locks, and the
  import lock (IMP_LOCK) at the start of py_os_after_fork_child
- Remove unlock() calls from _at_fork_reinit — unlock_slow()
  would attempt to unpark stale waiters from dead threads
- Replace try_lock() with lock() in after_fork_child now that
  locks are guaranteed to be in unlocked state

Note that this does not resolve any data deadlock issue