Fix Voice Fixer model loading and improve download speed
Technical details:
- Fix vocoder searching in wrong directory path (synthesis_module/44100 vs root)
- Move Voice Fixer node to correct category alongside audio processing nodes
- Add HuggingFace CLI download support to UnifiedDownloader for faster downloads
- Create HuggingFace mirror for Voice Fixer models (faster than Zenodo)
- Add UTF-8 encoding support for Windows compatibility
- Support both modern 'hf download' and legacy 'huggingface-cli download' commands
diodiogod
717a7afaf5