yakhyo/uniface ? reverse-engineered prompt

Build me a Python library called UniFace that gives me one simple API for face analysis. I want it to handle face detection, alignment, landmarks, recognition, tracking, parsing, gaze, head pose, age and gender, emotion, anti spoofing, face quality, anonymization, and portrait matting from images and video frames.

Please make it easy to use with a clean high level analyzer, plus separate modules for the different tasks, and include working examples and basic tests. It should automatically download and cache the needed models the first time they are used, and support CPU on Apple Silicon as well as NVIDIA GPU if available. Use current best practices for ONNX based inference, and look up any model or library docs online if you need to.

Also include a simple face embedding search store so I can compare faces quickly across a collection of images. Make the README examples actually runnable, and keep the setup straightforward for someone who just wants to pip install it and try detection or full face analysis right away.