fix deadlock on fast init deinit of filters

[summeroff]

Summary

• Fix deadlock in nvidia_audio_destroy when NVIDIA audio filters are rapidly created and destroyed (e.g. during automated tests that iterate through all filter types)

Problem

nvidia_audio_destroy locks ng->nvafx_mutex and then calls pthread_join on the init thread. The init thread (nvidia_audio_initialize) also needs to acquire ng->nvafx_mutex as its first step. When deferred destruction
runs before the init thread has acquired the mutex, this creates a deadlock:

• Destroy thread: holds nvafx_mutex, waits on pthread_join
• Init thread: waits on nvafx_mutex, never finishes

This cascades to nvidia_afx_initializer_mutex (the global mutex shared across all NVIDIA audio filter instances), blocking initialization of any subsequent NVIDIA audio filters and stalling obs_wait_for_destroy_queue.

Fix

Move pthread_join before the mutex lock in nvidia_audio_destroy. The init thread runs to completion first (releasing all locks), then destroy safely locks the mutex for cleanup.

Types of changes

• Bug fix (non-breaking change which fixes an issue)

Checklist:

• My code has been run through clang-format.
• I have read the contributing document.
• My code is not on the streamlabs branch.
• The code has been tested.
• All commit messages are properly formatted and commits squashed where appropriate.
• I have included updates to all appropriate documentation.