perf: add torch.compile, faster-whisper, and xFormers optimizations

perf: optimize gRPC server for lower latency

fix: use relative path for grpc imports in test client

feat: add temporal smoothing to reduce lip trembling in avatar output

feat: add MuseTalk gRPC streaming server for real-time lip-sync

feat: Add complete microservices architecture for speech-to-speech avatar system

fix: use get_detections_for_batch for face detection

fix: update fast_engine.py with Robin's working config

feat: add MuseTalk fast_engine integration and model download script

feat: add LiveKit WebRTC streaming support

feat: add webrtc_stream module for WebRTC video streaming

fix: handle webrtc_manager None in all places

fix: handle missing fast_engine and webrtc_stream modules gracefully

Initial commit: dumont-talker speech-to-speech avatar