sound: track QEMU output voices separately

sound: track QEMU output voices separately

Keep every QEMU output voice on a small linked list instead of
replacing a single global pointer. Pass the voice through the audio
callback so each active stream reads its own buffer.

Update all active QEMU voice volumes, only run QEMU board audio when a
voice is active, clamp writes to the sample buffer, and return no input
data until the capture path has a real backend.