[FEATURE] Adding ISNet as new model

**Is your feature request related to a problem? Please describe.**

I have been using ISNET quite a lot these days especially with a frozen GT encoder and found good results, would love to add it to the repo since its very good for segmentation

The issue names no files, tests, entry points, implementation details, or weight sources. Start by reviewing how existing image models are integrated and tested in the repository, then determine the required ISNet architecture, pretrained weights, registration, and segmentation behavior. Done means ISNet is available through the repository's normal model APIs with appropriate tests and documented weights.

Written by the indexing model from the issue text.