Add voice cloning service and update API for voice generation

Add voice cloning endpoint and XTTS model integration

Refactor: Hide model loading, focus on training pipeline

Initial HF Spaces deployment - downloads models at runtime