configurable augmentations

[mattersoflight]

A key hyperparameter of our training process is the choice of augmentations. At this point, the augmentations are hard coded. @ziw-liu how can we make the augmentations configurable via pytorch lightning config? A useful side-effect is that augmentations that go into each computational experiment will be automatically documented.

[ziw-liu]

I think this is possible with dependency injection, the same way as we planned for the models in #42 (comment).