UMAP dataframe distance metrics for mixed type categorical data

Nobody has claimed this yet.

I recently came across [this paper](https://www.ncbi.nlm.nih.gov/pubmed/30908240) and it seems potentially very useful for the work going on in the `dataframe_umap` branch.

To summarize, the authors provide an entropy-based distance metric that can take any kind of categorical data (nominal *and* ordinal), with appropriate weighting that balances the relative information content for each attribute/column in your dataframe. I found the paper was written clearly: the key result is equation 15, and they provide a suggested algorithm structure too. Is this something you might consider for a PR?

**Paper:**
Zhang Y, Cheung YM, Tan KC. A Unified Entropy-Based Distance Metric for Ordinal-and-Nominal-Attribute Data Clustering. IEEE Trans Neural Netw Learn Syst. 2019 Mar 19. doi: 10.1109/TNNLS.2019.2899381.
https://www.ncbi.nlm.nih.gov/pubmed/30908240

Related previous discussions:
https://github.com/lmcinnes/umap/issues/58
https://github.com/lmcinnes/umap/issues/206

Start by reading the linked paper, especially equation 15 and its suggested algorithm, then review the dataframe_umap branch and the related discussions in issues 58 and 206. A complete result would be a scoped, reviewed PR adding the entropy-based metric for mixed nominal and ordinal dataframe data, with its expected behavior agreed beforehand.

Written by the indexing model from the issue text.