refactor: strip to inference-only, add Docker support, fix WebSocket and UI

feat: add LiveAvatar streaming server with avatar management

feat: update download_weights.bat (#372)

fix: convert all audio to WAV 16kHz PCM before processing (#379)

fix: ensure upper bond does not go below zero in landmark extraction (#329)

fix: preprocess import bug (#345)

fix: use torch.no_grad() in inference to prevent excessive memory usage (~30GB) with inference (#349)

fix: download_weights.sh (#318)

feat: windows infer & gradio (#312)

Update audio_processor.py

docs: update readme

feat: data preprocessing and training (#294)

fix: floor (#293)

fix: bug in infer script (#290)

feat: real-time infer (#286)