Reuse Keras models optimization

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

Change ensemble size from 8 to 12

Bump keras version, set mhcflurry version to 0.9.1

Small updates

* Add more experiments to models_class1_experiments
* Add `--quantitative-only` flag to training script
* Add `--no-throw` flag to prediction script

Bump travis python 3 version to 3.6

Add helper functions for potential mhcflurry wrappers

MHCflurry 0.9

Switch downloads_command to use mkstemp instead of NamedTemporaryFile for windows support. Fixes #79