feat(music-separation): implement real Demucs-based separation with music preservation

[cto-new]

Summary

This PR implements real Demucs-based music separation and a music preservation workflow to remix dubbed voice with original music while preserving audio quality. It also introduces metrics and tests to validate separation and preservation.

Details

• Replaced mocked separation with real Demucs-based model supporting vocals, drums, bass, and other stems
• Added MusicPreservationPipeline to remix dubbed voice with original music while preserving timbre and dynamics
• Implemented MetricsCalculator with MusicPreservationMetrics and SeparationQualityMetrics to quantify preservation and separation quality
• Introduced MusicSeparationOrchestrator to coordinate separation, preservation, and validation
• Expanded audio utilities for multi-format I/O, resampling, and normalization
• Added tests (unit and integration) and updated README with usage and acceptance criteria

Warning: Task VM test is not passing, cto.new will perform much better if you fix the setup