macOS: auto-select Whisper MPS device and add macOS setup instructions

[mpaepper]

Motivation

• Make Vibevoice runnable on macOS (Apple Silicon) by selecting a sensible Whisper runtime and documenting macOS-specific steps.
• Reduce manual edits needed to run on MacBooks (MPS support, PortAudio via Homebrew, and pynput accessibility permissions).

Description

• Add OS-aware runtime selection in src/vibevoice/server.py that picks mps/float16 on macOS and cuda/float16 elsewhere, with environment overrides WHISPER_DEVICE and WHISPER_COMPUTE_TYPE.
• Initialize the WhisperModel using the chosen runtime (WHISPER_DEVICE, WHISPER_COMPUTE_TYPE).
• Update README.md with a macOS setup section including brew install portaudio, pip install -r requirements.txt, pynput Accessibility/Accessibility permissions guidance, example run command, and environment variable override examples.
• Files changed: src/vibevoice/server.py, README.md.

Testing

• Ran python -m compileall src/vibevoice/server.py src/vibevoice/cli.py, which completed successfully.
• No other automated tests were run in this change.

---

Codex Task