feat(audio/linux): host-owned stream sink decouples capture from hardware-sink churn

The Linux desktop-audio capture stream now registers itself as an Audio/Sink
node ("Punktfunk Stream Speaker", default-on, PUNKTFUNK_STREAM_SINK=0 =
legacy escape hatch) and claims the configured default sink for the duration
of a session (saved and restored around it, refcounted across concurrent
sessions, crash-stale claims degrade to automatic election). Host apps play
directly into the capture stream, so the capture link no longer depends on
any hardware sink.

Root cause this fixes (live-diagnosed on a bazzite/LG-TV host): gamescope
modesets drop the NVIDIA HDMI audio endpoint, WirePlumber ping-pongs the
default sink HDMI<->auto_null ~8x/s, and the old monitor-follower relinked
its capture on every flip - Paused/renegotiate/Streaming storms (~1300 log
lines/min) heard as crackle on the client.

Also fixes a latent liveness bug in both modes: the capture thread had no
core-error listener, so a PipeWire daemon restart mid-session left a zombie
thread returning quiet-sink empty chunks forever (same class as the historic
virtual-mic death bug). Now the thread exits and sessions reopen with backoff.

Bonus: the sink advertises the session's true channel count, so games can
produce real 5.1/7.1 even when local hardware is stereo. New AudioCapturer::
idle() hook releases the routing claim when a capturer is parked between
sessions; drain() re-claims on reuse.

Verified: cargo clippy -D warnings + 295 tests green on Linux (.21).
On-glass validation on the bazzite host pending.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

crates/punktfunk-host/src/gamestream/audio.rs

@@ -297,6 +297,7 @@ fn run(
         None => audio::open_audio_capture(want).context("open audio capture")?,
     };
     let result = audio_body(&mut *cap, &sock, gcm_key, rikeyid, params, running);
+    cap.idle(); // parked between sessions — release the routing claim (Linux stream sink)
     *audio_cap.lock().unwrap() = Some(cap);
     result
 }