feat: backend-specific ArcFace thresholds (onnx cosine=0.50, deepface cosine=0.68)

fix(ui): make detector name dynamic per backend

refactor: switch default backend to onnx, split per-backend Dockerfiles

refactor: clean up app.py, slim API responses, document ONNX backend

fix: add tf-keras dependency required by retinaface with TensorFlow 2.21+

fix: change default FACE_BACKEND from onnx to deepface for better HF Space performance

fix: remove tensorflow-cpu pin to resolve ml_dtypes conflict with onnx

fix: add build-essential for insightface Cython compilation

fix: sync Dockerfile with HF Space (python 3.10, libgl1, libhdf5-dev)

feat: change default FACE_BACKEND to onnx

feat: add ONNX backend, timing display, and no-face detection message

fix: add tensorflow-cpu explicitly to fix deployment on Debian Trixie

feat: initial face detection and compare web app using RetinaFace + ArcFace

Initial commit