docs: publish desktop audio with PocketStation - #805

Summary

Add a runnable Python example that publishes one local audio source to a LiveKit room:

• a selected desktop application;
• the complete desktop output mix; or
• an explicitly selected microphone.

PocketStation owns local capture and delivers source-aware frames outside the operating-system callback. The example downmixes those frames to mono PCM16 and publishes them through LiveKit AudioSource. Application and system audio are marked as screen-share audio; microphone capture is never enabled automatically.

The example uses released packages only: pocketstation==0.1.4, livekit==1.1.17, and livekit-api==1.2.1.

Verification

Commit: 6632785e1c195ebb5879e2cc97171f5bfeed9238

uvx ruff check examples/desktop_audio/publish.py examples/desktop_audio/tests/test_publish.py
uvx ruff format --check examples/desktop_audio/publish.py examples/desktop_audio/tests/test_publish.py
python -m unittest discover examples/desktop_audio/tests -v
mypy --strict examples/desktop_audio/publish.py

A clean Python 3.11 environment installed all three packages from PyPI and passed the conversion test.

I also ran a local LiveKit 1.13.3 server with separate publisher and receiver participants on macOS:

• selected QuickTime Player: 1,126 PocketStation frames submitted; receiver observed 643 frames, 308,640 samples, 30,670 nonzero samples, peak 32,767;
• system audio: 1,126 PocketStation frames submitted; receiver observed 627 frames, 300,960 samples, 72,659 nonzero samples, peak 32,768.

This proves native macOS capture and same-host LiveKit delivery. It does not qualify WAN/TURN behavior. The microphone option was not opened during this run; the README directs interactive voice applications to LiveKit PlatformAudio for AEC, noise suppression, and gain control.