Add progress bar to RandomForest

classify.RandomForest uses the VIGRA library, which is written in C++ and there's not a straightforward way to have a progress bar. Instead, we could train a 1-tree random forest before beginning the full training to see how long it takes to train a tree.

On one test dataset, indications are that this would provide a good estimate of total time (i.e., initial overhead and other factors won't mess things up), since we get the following times for training 1, 2, and 3 trees:

1 tree: 118.812502146 seconds
2 trees: 236.317131042 seconds
3 trees: 351.319090128 seconds

Each successive tree is very close to a multiple of the 1-tree training time. The progress bar could then be based on time.