Transformer chaos containment

[texasmichelle]

Before more transformer-based language models are added to this repo, let's pay down some debt on the ones we have:

• [ ] #315 Various BERT cleanup
• [x] #480 Use Epochs as a data loader in CoLA
• [ ] #488 Data source resolution in CoLA
• [ ] #489 Use common file extraction functions in CoLA
• [ ] #432 Disentangle CoLA and BERT
• [ ] #433 Verify BERT variants (ALBERT and RoBERTA are untested)
• [ ] #434 GPT-2 and BERT alignment

This will make additional models easier to add and maintain.