nfaggian/growcut ? reverse-engineered prompt

Build me a small Python package for image segmentation using the GrowCut algorithm.

I want it to take an image and a few user marked seed points, then grow those regions until the image is split into labeled segments. It should work on normal grayscale or color images, feel like a reusable library, and include a simple demo notebook showing how to run it on an example image and visualize the result.

Please make sure it has tests, installs cleanly, and works with the usual scientific Python stack. If you need to check current docs or a better implementation detail, look them up online. Keep the code easy to read so I can understand how the algorithm is working and tweak it later.