feat(voice-cloning): real ECAPA-TDNN embeddings, enrollment, and multi-speaker support with PostgreSQL

[cto-new]

Summary

Introduce real speaker embeddings via ECAPA-TDNN, enabling speaker enrollment and multi-speaker support for cross-language voice cloning. Data is persisted in PostgreSQL with embeddings, enrollments and voice characteristics.

Details

• Replace mocked embeddings with real ECAPA-TDNN embeddings and add a speaker enrollment pipeline.
• Implement multi-speaker detection/diarization with segmentation, clustering, and per-speaker audio extraction.
• Persist embeddings, enrollments, and voice characteristics in PostgreSQL; enable reuse of embeddings across languages.
• Add VoiceAnalyzer to extract pitch, speaking rate, and spectral characteristics for richer speaker profiling.
• Provide scaffolding for cross-language voice cloning (TTS integration placeholder) and comprehensive tests and docs updates.

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