feat: make Stream implement Send on all hosts

[roderickvd]

Stream can now be moved between threads safely on macOS.

• Replaced Rc<RefCell<>> with Arc<Mutex<>> for thread-safe access to stream state
• Extracted device disconnection handling into a separate DisconnectManager to isolate non-Send AudioObjectPropertyListener
• Removed unnecessary unsafe impl Send implementations that are now automatically derived

The mutex is only locked during play() and pause() operations, so performance impact is negligible. The audio callback is completely unaffected.

[dvdsk]

I suspect we can not do this without spawning a separate thread to keep AudioObjectPropertyListener in. Maybe its perfectly safe maybe its not. There is not enough documentation from Apple to judge that.

Lets give Apple users an extra thread doing nothing but keep that object on the same thread.... We should name the thread: apple_should_document_their_audio_api \s

[roderickvd]

@dvdsk good catch. Let's be more conservative. I've reworked this to use a dedicated thread.

While I was at it, I also killed the Clone impl on Stream. That's just confusing, and ALSA and WASAPI don't have it either. Should be much cleaner now.

[roderickvd]

This should be in a good state now - thanks @dvdsk for the necessary reviews. I'll rebase and if you agree, we can merge.

[edwloef]

It would be nice if various backends could be tested on their send-ness, to make sure this doesn't accidentally regress. Something on the lines of

const fn assert_send<T: Send>() {}
const _: () = assert_send::<Stream>();

maybe, so that a non-send stream is a compile fail?

[roderickvd]

It would be nice if various backends could be tested on their send-ness, to make sure this doesn't accidentally regress. Something on the lines of

const fn assert_send<T: Send>() {}
const _: () = assert_send::<Stream>();

Adding that in 1d117d3 was pretty revealing: WASAPI and WASM streams aren't Send, either. I'm gonna see if that's easy enough to fix to include with this PR.

[edwloef]

It would be nice if various backends could be tested on their send-ness, to make sure this doesn't accidentally regress. Something on the lines of

const fn assert_send<T: Send>() {}
const _: () = assert_send::<Stream>();

Adding that in 1d117d3 was pretty revealing: WASAPI and WASM streams aren't Send, either. I'm gonna see if that's easy enough to fix to include with this PR.

On WASM this is probably fine since there's no real concept of threading as far as I'm aware, but the WASAPI stream being Send would be nice.

[roderickvd]

On WASM this is probably fine since there's no real concept of threading as far as I'm aware, but the WASAPI stream being Send would be nice.

You're right. Turned out to be easy for both, so added it here.

[roderickvd]

Merging this now because I want to start preparing the next release. Please create an issue if something's up with it.