Fix volume control threading and cleanup bugs

[teancom]

Summary

These fixes address several related threading issues in the volume control subsystem. They all stem from the same root cause — polling threads not being properly lifecycle-managed — so they're bundled together rather than split into separate PRs. BUT! If you prefer separate PRs, I can absolutely do that. Trying to balance 'number of PRs to review' with 'size of PR to review', the eternal struggle.

Fix 1: Windows COM initialization on polling thread

Bug: The volume polling thread calls COM methods (GetMasterVolumeLevelScalar, GetMute) but never called CoInitializeEx on that thread. COM requires per-thread initialization. This could cause CO_E_NOTINITIALIZED failures, meaning external volume changes would silently go undetected on some Windows configurations.

Fix: Added CoInitializeEx(None, COINIT_MULTITHREADED) at the start of the polling thread and CoUninitialize() after the loop exits.

Fix 2: Windows Drop ordering

Bug: The existing Drop impl called CoUninitialize on the creating thread while the polling thread was still running and actively calling COM methods through its Arc. This is a use-after-free.

Fix: Reordered Drop to: set stop flag → join polling thread (which does its own CoUninitialize) → drop the endpoint volume → CoUninitialize on the creating thread. This matches the principle that no COM calls should happen after uninitialization.

Fix 3: macOS missing Drop impl

Bug: MacOSVolumeControl had no Drop impl at all. Dropping the controller abandoned the polling thread, which would run until process exit. Repeated start/stop cycles would accumulate leaked threads.

Fix: Added a Drop impl using the same Arc stop flag pattern as Windows — set the flag, join the thread.

Fix 4: VOLUME_CONTROLLER never cleared in stop()

Bug: stop() cleared SHUTDOWN_TX, COMMAND_TX, and SENDSPIN_CLIENT, but never VOLUME_CONTROLLER. This meant the platform-specific Drop impls (including the new ones from fixes 2 and 3) never ran during normal shutdown — only at process exit, if at all.

Fix: Now stop() clears VOLUME_CONTROLLER as its first action, triggering Drop and joining the polling thread before the rest of shutdown proceeds.

Fix 5: set_change_callback thread leak on repeated calls

Bug: On both Windows and macOS, calling set_change_callback a second time would overwrite the stored thread handle without stopping the previous polling thread. The old thread would run forever with a lost JoinHandle.

Fix: Now both platforms stop and join the existing thread before spawning a new one, using a fresh stop flag so the new thread isn't born already signaled.

[marcelveldt]

Thanks! One more thing that we should probably implement with regards to volume control is the option to choose between hardware or software volume control but that is for later.

[marcelveldt]

Thanks @teancom !

[teancom]

I'm taking a poke at the software volume control thing, no promises it'll actually be usable. 😅 Is this mergable as-is or do you want me to submit the bug fixes and the feature together?

[marcelveldt]

Fine like this, thanks @teancom !