AD index metric warning

[stewarthe6]

Some metric/feature combinations are inappropriate and will calls AD index calculations to fail. For example, Jaccard distance with RDKit features results in a pairwise distance matrix that is all zeros. There should be a warning when this happens.

A check can be inserted here.

[stewarthe6]

There should be a warning when it's a bad idea like, euclidean on fingerprints and an error for things like jaccard and rdkit features.

[stewarthe6]

Maybe update to use Facebook's faiss function to do pairwise_distances for large datasets.
https://github.com/facebookresearch/faiss

Should replace default error value of 1e-6 with NaN to be explicit about a failure.