Feature/rank features group

[VladimirShitov]

PR Checklist

• This comment contains a description of changes (with reason)
• Referenced issue is linked: Add more statistical tests for comparing features #532
• If you've fixed a bug or added code that should be tested, add tests!
• Documentation in docs is updated

Description of changes
Previously, for non-numerical features the standard statistical test was run by ep.tl.rank_features_groups (e.g. Wilcoxon rank sum test). This PR adds functionality to run statistical tests specifically developed for categorical features (e.g. Chi-square test).

Technical details

• The same approach that in scanpy.tl.rank_genes_groups is used. E.g., when the reference is set to "rest", for each subgroup of groupby, the composition of a categorical variable is compared to the composition in all other groups mixed together. This is not a common approach, I would say, but it is consistent with scanpy, which is used for numerical features.
• The default test is G-test, which is similar to the Chi-square test but should work better for groups with a small expected number of observations.
• P-values should be treated carefully. I would only use them for ranking marker features, and re-run statistical analysis in a conventional way to test your hypotheses.

[VladimirShitov]

Note: function parameters are not validated as extensively as in scanpy. I might add this in coming commits

[Zethson]

You're a legend @VladimirShitov !

1. Left lots of minor comments
2. A file named _utils is usually a slight code smell because ideally ever piece of code should have a clear purpose. I would actually move ALL of the feature_ranks code including this into a _feature_ranks_groups.py. What do you think? IMO we have lots of customization now and this would make sense.

Thank you so much.

[VladimirShitov]

Thank you for your comments, Lukas! Please, check the discussion on renaming datatypes.py above. Everything else is fixed