add option to control min_samples_leaf for decision trees and random forest

[jowagner]

Currently, a few decision tree sizes are hard-coded in train.py. Add an option.