Add utility for training to dirty image and integrate into cross-val

[jeffjennings]

• Adds a function to train against a dirty image.
• Optionally integrates this function into a cross-validation loop.
  • Training against a dirty image is run once at the start of cross-val, with the resulting state dictionary saved. The state dict is then loaded to update each new model at the start of the training loop for each k-fold.
• Adds a test of the function.

[iancze]

Looks good, thanks. It would be helpful to have documentation (a tutorial, or other narrative text) explaining how one might start using these tools in an example workflow. Perhaps for HD 143006 part 3, or the IM Lup example.

[jeffjennings]

Great, and good point - the HD 143006 part 3 tutorial sounds like a nice place. I've added a note in #63.