feat(tts): improve voice picker UX and expand voice lists

Current voice picker is a flat popover list, doesn't scale well for providers with many voices (MiniMax has 327, Azure has hundreds by region).

Planned improvements:
- Add search/filter to voice picker popover
- Group voices by language with collapsible sections (hot languages expanded by default)
- Dynamic voice list fetching for providers that support it (MiniMax /v1/get_voice, Azure /cognitiveservices/voices/list)
- Expand static voice lists: MiniMax (5 -> 30+ common), GLM (add missing `female` voice)
- Verify ElevenLabs voice ID format (names vs IDs)
- Extract large voice lists to separate files (e.g. lib/audio/voices/minimax-voices.ts)