Scale up training and testing; compare runtimes

[matsen]

1. Update Snakefile for training/testing on multiple datasets and aggregating
   output:

• train on all combinations of training and testing data that is input in
  config. For input data [train1, train2] and [test1, test2] we currently
  only have two runs: (train on train1 and test on test1) and (train on
  train2 and test on test2). We want to extend this to all possible
  combinations of training and testing data

2. Compare training times on CPU with those on GPU
   • try both the ermine GPU and FH cluster
     GPUs