Fix: close PIL images and PDF handles; fix temp-file cleanup in backend

Fix: cache YOLO model to prevent reloading weights on every logo detection call

Fix: cache transformer model to prevent reload on every table detection call

Final

Fix: lazy import unstract, add to requirements — OCR skips gracefully without API key

deployement started

added DB

Fixed different cbse template and contrast issues

Modified the flow and added college marksheet processing

Initial commit with Git LFS tracking