v0.2.0

• In this release a problem class was created in order to extract loss function and metrics from a compiled tf.keras.Model.
• This way of doing things is much more compatible with keras workflows than the old way of doing things in hessianlearn.
• Additionally the new kerasModelProblem class can handle multi-input and multi-output networks.
• This version creates more flexibility with passing optimizers into the model wrapper that handles training.