Redo `_preprocess_df()` when sklearn ColumnTransformer has inverse_transform method

[tsrobinson]

sklearn's ColumnTransformer has good functionality for mixed-data pre-processing, and would tidy up some of our code. Currently sklearn lacks inverse transform for this specific Transformer, and although requested here scikit-learn/scikit-learn#11463 and fix proposed here scikit-learn/scikit-learn#11639 does not seem to be implemented yet.

The basic workflow would be to:

1. Detect datatypes
2. Build column transformer CT with numeric and categorical encoders
3. Run SyGNet
4. Inverse transform generated data using CT.inverse_transform()

Assuming this method is implemented at some point, we should revise our function.

[tsrobinson]

Still waiting...