Changes:
- Remove all volume management (use VolumeNode in pipeline instead)
- Detect hardware format (I16/U16/F32) at startup
- Accept all AudioChunk formats (I16/I24/I32/F32/F64) as input
- Use optimized SIMD functions from dsp::int_float module
- SharedBuffer stores raw AudioChunk + intermediate F32 buffer
- Callbacks adapted to hardware format with proper conversion
Architecture:
1. AudioChunk pushed to SharedBuffer
2. Lazy conversion to F32 interleaved using dsp functions
3. Callback converts F32 → hardware format (I16/U16) if needed
Benefits:
- SIMD optimized conversions (dsp module)
- Clean separation of concerns (volume in VolumeNode)
- Hardware format detection (use native format when possible)
- Flexible input (accepts any AudioChunk type)
Note: Requires ALSA (libasound2-dev) on Linux for compilation
30d30739bd