serengil/retinaface — reverse-engineered prompt

Build me a Python package for face detection called RetinaFace. I want it to take an image path, find every face in the photo, and return a clean result with a confidence score, the face box coordinates, and facial landmarks like the eyes, nose, and mouth corners. It should work well even when there are multiple people in the image.

I also want a simple way to extract the detected faces as image crops, with an option to align them based on the eye positions so they are ready for face recognition pipelines later. Keep the API very easy to use, something like one function for detection and one for extracting aligned faces.

Please make it feel like a real reusable library, installable with pip, with a small example in the docs and a couple of tests so someone can verify it works quickly. If face recognition itself is out of scope, that is fine, just make this detector and alignment part solid. Look up current docs online if you need to.