v2.0.0: Silero TTS — production-ready multilingual TTS

Replaced XTTS-v2 with Silero Models:
- 50+ speakers across 15+ languages (Kazakh, Russian, English, Tatar, Uzbek, etc.)
- GPU-accelerated: ~108 req/s, <50ms avg latency
- Auto-chunking: texts up to 10000 chars split by sentences automatically
- Async inference lock for thread safety
- Input validation (speaker, sample_rate, speed, text length)
- Generation timeout (504 on exceed)
- Optional API Key auth (Bearer token)
- OpenAI-compatible API (/v1/audio/speech)
- Pinned dependencies and Silero hub version
- Health check endpoint