Dynamic TCR sets

[matsen]

From @krdav :

I see that you maintain my "macro like" globals defining V/J genes. They have been annoying me from the beginning because I wanted it to be dynamically defined by the dataset. However, make a sufficiently small (or single subject) hold-out set and it will surely not contain all the genes of the full dataset. It also causes compatibility issues when predicting on an completely different dataset, not used for training.
That being said if/when this is moved from notebooks towards command line scripts it may be worth defining allowed V/J genes (as well as similar info like HLA type) in an input file. This could have a default in the repo to ease use.