Add additional dimensionality reduction methods

[enjalot]

It would be interesting to be able to swap PHATE, t-SNE or even PCA for UMAP for dimensionality reduction.

I'm a big believer in UMAP being the best dimensionality reduction algorithm, but there are valid cases for other approaches (even if its just to see for yourself that UMAP is the best 😆 )

I'm imagining step 2 becomes "Projection" instead of "UMAP" and then there is a toggle to switch between umap, PCA, tsne and phate.
The Setup UI would then need to show slightly different parameters.

I'm thinking this would be implemented as a new phater.py script similar to umapper.py that does PHATE and outputs phate-001.parquet etc.
Then when you cluster you could pass in either a umap-001 or phate-001. Perhaps renaming the umaps folder to projections and allowing either umap-* or phate-* would be the way to go, as the outputs will be essentially identical, but recording the inputs will be important.
This would also mean having references to umap_id in cluster-.json and scope-.json be updated to projection_id

[enjalot]

Another one to look into: https://github.com/scikit-tda/kepler-mapper