Pytorch refactor

[jatkinson1000]

Some of the pytorch in this repository is written in unconventional or inefficient ways that can make it hard to follow, or run at suboptimal levels of performance.

As part of the more detailed cleanup the working code needs looking at and refactoring.

This will involve branching off the documentation branch (to leverage the current docstring and linting improvements) and then merging back into main after #17 has closed #7 to try and minimise conflicts.

It should also not be merged back until #6 has been closed so that tests can be run to ensure behaviour is unchanged following refactoring.

The following approach is suggested:

• create a new module FullyCNN.py or similar to house model, and perform updates in this.
• refactor models1.py (and associated code in models/*) Replace models1 usage with fully_conv_net #43
• Note of any dead files or code to be removed later at merge
• Check for conflicts with Add code Documentation and linting #17 after merged
• Passes tests developed in Add testing framework #6

[mondus]

Cover in other issues #43 and #40