Fixing ALSA underrun death spiral and PulseAudio buffer clamping

[f1nalspace]

Issues

• Breaks sound server playback (PireWire, PulseAudio, ALSA)
• PipeWire/PulseAudio audio buffer underrun
• Playing random noises not recovering without a full service restart

Bugfixes

ALSA:

• After snd_pcm_set_params, override sw_params to set start_threshold to period_size instead of buffer_size. The default causes ~200ms of silence after every underrun recovery, during which PipeWire fires again and triggers another underrun — a death spiral.
• Set silence_size=buffer_size so ALSA auto-fills silence during brief packet gaps, preventing XRUN before snd_pcm_writei can be called.
• Add snd_pcm_drop() before snd_pcm_close() to prevent hang on close.
• Use continue instead of return 0 after snd_pcm_recover so the current packet is retried rather than dropped.

PulseAudio:

• On format change, recalculate maxlength with the updated sample spec before recalculating tlength. Previously maxlength was set at init with the 44100 S16LE placeholder and never updated; after switching to 48kHz S32LE, tlength > maxlength, so PA silently clamped the effective buffer to ~46ms instead of the requested latency — causing the same underrun spiral.

[f1nalspace]

The problem was gone for a while, but now even with all those fixes, the noises are back - especially when i play a video on my host and install stuff in the VM that plays system sounds.