Consider making audio abstraction push-based

[yupferris]

This should help to "black box" the emulator code in general, and removes the need for any threading code etc. If the emulator's bootstrapping wants to switch between time sources (for instance, if the audio output gets muted, you might want to drive the emulator using a timer based on the desired framerate instead) this would make it much easier.

Tell the emu to run x time units, and take whatever output it may have produced from its output buffers. This makes sense to me :)