fix(capture): restore proven-clean wallclock audio (match de509c6 baseline)

Removing wallclock made A/V length drift far worse (audio 11.8% long). The
known-clean config used wallclock + master aresample=async=1; the leading
silence is a standby backlog artifact addressed by the bridge live-edge flush +
record-start audio FIFO drain, not by changing the timestamp source.

services/capture/src/capture-manager.js

@@ -720,15 +720,11 @@ class CaptureManager {
           '-video_size', fcSize,
           '-framerate',  fcFps,
           '-i', 'pipe:0',
-          // Audio FIFO → ffmpeg input 1. The bridge flushes its slot backlog to
+          // Audio FIFO → ffmpeg input 1. Wallclock timestamps + master
-          // the live edge on reader-attach, so the FIFO delivers live audio in
+          // aresample=async=1 is the proven-clean A/V config; both inputs must
-          // lockstep with the video. We DERIVE audio PTS from the s16le sample
+          // start from the live edge (see the fc_pipe + audio flush at record
-          // count starting at 0 — the SAME origin as the video's frame-0 PTS —
+          // start) so aresample has minimal correction to do.
-          // rather than from wall-clock arrival. Wall-clock stamped the first
+          '-use_wallclock_as_timestamps', '1',
-          // audio chunk at a time offset from video frame 0, and the master
-          // aresample then PADDED ~2.5s of leading silence to align them. With
-          // sample-count PTS both streams share one origin → no pad, no leading
-          // silence, length locked.
           '-thread_queue_size', '512',
           '-f', 's16le',
           '-ar', '48000',