Audio API Improvements

[revmischa]

The audio API changed quite a bit in SDL3.

I tried rearranging some things including how to open streams with callbacks and devices. I think the callback form is still pretty important for opening recording streams, since you need a callback to be called when audio data is received. Filling the audio buffer for playback is more reasonable without a callback.

Need to re-think how locking should work, what methods belong on AudioSubsystem vs. AudioDevice vs. AudioStream etc.

Open to suggestions/ideas.

I'm trying to make use of the recording stream API in a real-world example here: projectM-visualizer/frontend-sdl-rust#10 and trying to adjust the API to suit my needs.