[BUG] frozen ai deps for dirty-cat ci failures

[lmeyerov]

Commit https://github.com/graphistry/pygraphistry/commit/e1be5e46d25d6224ed80f1b5c5012470c9b08212 requires, for umap/ai installs: py3.8+ (vs 3.6), scikit-learn 1.0+ (vs 0.2x), and froze dirty-cat to 0.2.0 (disallows 0.2.1)

This is to work around CI fail FAILED graphistry/tests/test_umap_utils.py::TestUMAPFitTransform::test_allclose_fit_transform_on_same_data

This a problem in a few ways:

• we cannot use latest dirty cat
• scikit-learn / py3.8 are relatively new, so annoying as required deps

Next step may be to file a minimal reproducible example with dirty-cat on their repo? Ex: Show something that works in the above setup and fails in dirty-cat 0.2.1 Ex: Show something that works in the above setup and fails in dirty-cat 0.2.0 with scikit-learn 0.2x (which they aim to support afaict)