Explore widgets fail on unexpected datetypes

[diehlbw]

Problem Summary

When our exploration widgets (found using ExploreCohortEvaluation) try to calculate performance metrics a first step is to do some splitting. It works well for categoricals, but then defaults to an assumption of continuous. This works well for actual numerical data but throws an unhelpful error if this is not true.

Impact

Looks broken, and is non-obvious what the cause is

Possible Solution

Best solution - add handling for more datatypes
Patch fix - catch TypeErrors and provide a useful message about the functional limitation