Performance enhancement for faster start time: reuse compiled networks even if their regularizations differ. This is safe because regularization should not affect predictions

Docs improvememnts, more flexible Keras dependency, drop typechecks dependency, remove some unused functions

* First cut of supporting quantiles ("percent ranks") per allele. Closes #87.
* Some refactoring of amino_acid.py

This builds on #114

This should give a substantial prediction speed improvement as identified by @rohanpai

* Add borrow_cached_network() method to Class1NeuralNetwork
* Use cached models at in Class1NeuralNetwork.predict
* Set names for all neural network layers so the model JSON is identical for identical architectures