#140 Add chunked / append audio (A;Load) so long samples can play before full transfer and interleave with video frames

[Rob Swindell]

The SyncTERM:A;* audio model loads a complete sample: C;S stores the file,
A;Load;S=<slot> decodes the entire file (libsndfile) into a PCM patch, and
A;Queue plays it. That is ideal for short sound effects. For a long music
track (a full OGG/WAV) two things bite:

1. No play-before-complete. Playback cannot begin until the whole file has
   transferred and A;Load has decoded it — first-play latency = full transfer + full decode.
2. It stalls video. The door sends over one TCP stream. A music C;S is one
   contiguous APC, and because an APC (ESC _ … ESC \) cannot be nested, no
   frame APCs can be sent until it finishes — so a large first-time music upload
   freezes the display (the door drops frames) for the upload's duration.

On the wire:

  …frameN(ESC_…ESC\)  [ music C;S APC: ESC_ … … … ESC\ ]  frameN+k(ESC_…ESC\)…
                      └─ contiguous; frames N+1…N+k-1 dropped ─┘

(The doors already reduce the repeat cost with C;L upload-once-per-client plus
a render-once disk cache, so this mainly bites the first upload of each track —
but that freeze is user-visible.)

Request: a chunked / append audio transfer — e.g. A;Load accepting a
sample across multiple APCs (append-to-slot), with SyncTERM decoding
incrementally and starting playback once enough PCM is buffered. That gives:

• long music starting to play before the whole file has arrived, and
• the door emitting the sample in small pieces between frame APCs, so the
  transfer interleaves with video instead of blocking it.

A lighter first step: even an append/chunk store with no incremental decode
would let frames interleave (removing the freeze), with an optional
"start at N% buffered" to approximate streaming.

[Stephen James Hurd]

This is unlikely to ever happen. Neither the graphics nor audio APC functions are intended to be used for streaming. Further, nothing is stopping you from chunking it on your end and playing the samples back-to-back in the same channel... that's exactly why channels have queues.