CI tests

prep
adjust test bounds
figure out how to run "kpython" on ci
create my own docker images (one without auto_ml that we use for tests, one with the latest auto_ml)
    essentially, copy the kaggle one, include the stuff that i need included, and rip out all the extra stuff i don't so it downloads much more rapidly
figure out caching of docker images on ci??

ship the ci improvements

update documentation (train_categorical_ensemble and feature_learning/fl_data)
update version
upload to pip
cut a new release tag





