Network improvements - Githubissues

This PR contains various improvements for the network setup:

Make sure that test_and_train.py does not depend on anything Firedrake related.
Increase the batch size when running on GPU.
Use float32 on GPU and double on CPU.
Default to infinite patience on adapting the learning rate.
Default to infinite patience on early stopping.
Rename SimpleNet as SingleLayerFCNN.

jwallwork23 / nn_adapt