fix: prevent setting torch bridge in preprocess_video for numpy compatibility

fix: CPU-safe inference for HF free tier

Update: model.py and app.py to remove multiple instances of same methods and add quantization (f16) to reduce inference time

feat: add /predict_frames endpoint and frame-based preprocessing to FastAPI service

feat: add /predict_frames endpoint and frame-based preprocessing to FastAPI service

update: api calls and data preprocessing

fix: update Dockerfile and requirements for video processing dependencies; enhance preprocess_video function

add: implement SwinTClassifications model and video processing functions