Refactor audio pipeline for multi-client streaming and improved error handling

Refactor the audio pipeline to support multi-client streaming with new OggFlacStreamHandle and StreamingOggFlacSink.

- Replace DirectOggFlacSink with StreamingOggFlacSink in pipeline
- Update documentation and comments to reflect multi-client support
- Add detailed warning logs for buffer underruns and client disconnections
- Remove TimerBufferNode from pipeline as it's no longer needed
- Update connect() calls to subscribe() for new streaming behavior

This change enables multiple clients to subscribe to the same audio stream, with each subscription getting a live feed from the current point in time.

pmoaudio/src/nodes/timer_buffer_node.rs

@@ -135,6 +135,14 @@ impl TimerBufferNodeLogic {
                 self.buffer.len()
             );
 
+            if self.buffer.is_empty() && self.buffered_time_sec < self.capacity_sec * 0.1 {
+                tracing::warn!(
+                    "TimerBufferNode: buffer underrun at ts={:.3}s (capacity={:.1}s) — source too slow or stalled",
+                    segment.timestamp_sec,
+                    self.capacity_sec,
+                );
+            }
+
             send_to_children(std::any::type_name::<Self>(), output, segment).await?;
         }
         Ok(())