Relaxed Image validation limits for mobile photos

fix: add huggingface_hub to auto-resolve LFS pointers dynamically and fix load state

fix: set weights_only=False to allow PyTorch to load Swin Transformer weights

fix: update OpenCV docker dependency for newer python slim image

fix: add exact Huggingface Space Config

fix: deploy backend with LFS tracking for 310MB model weights

fix: reject non-skin images and fix false low-quality rejection

fix: move backend files into backend folder