Revert whisper small back to base — small too heavy for free tier

Switch whisper base->small for much better multilingual STT

Switch whisper tiny->base for better multilingual STT + beam_size=3 + VAD filter

Non-blocking processing threads + webm-to-wav for whisper + hallucination filter

Language param + posture every 2nd frame

Switch to multilingual whisper tiny model (99 languages)

Update live_processor.py - custom webcam component, no WebRTC

Update live_processor.py - unified live+video view, no separate tabs

Update live_processor.py - live session with Mamdani fuzzy fusion