Training refactoring

[crangelsmith]

As described in #296, simplifying the training script to make it clear what is happening.

• Moved model and optimiser configuration to its own files.
• Created function for gamma/beta annealing configuration, and moved it to its own file.
• Moved pass_batch functionally back to train.py and evaluation.py as it can be a bit confusing to see what is happening if is not explicit in the batch loop.
• Aesthetic changes to make train.py script shorter.
• Renamed variables for further clarity (e.g input_size to input_shape).
• Added missing doctrings.
• Updated readme with the latest configuration option and some aesthetic changes for readability (a lot more changes are needed to the README).

[crangelsmith]

@jolaem would it be ok to merge this PR?

[crangelsmith]

I'll merge this now as is stopping further progress.