EasyPreprocessor fails on categorical data as integers with missing values

Filling missing int or float values with ``dabl_missing`` breaks.
We might want to call ``clean`` on the data first (which converts the categories to string....), or have a separate ``OneHotEncoder`` for non-string categories. That would make the feature names harder to interpret because missingness will be encoded as and integer or float then.

Alternatively we can just error on dirty data?