Changed the covariates to be more clever with how they produce the different splits. In the future (not yet implemented) a clever GroupDifferentiator
could update the current score calculation based just on how many rows moved from one hand to the other. There were a few other changes as well;
TreeTrainer#growTree now accepts a Random as a parameter which is used throughout the entire growing process. This means it's now theoretically
possible to grow trees using a seed, so that results can be fully reproducible.
SplitRules are tested.
Most of the refactoring involved the creation of a Covariate class (one
instance per column); with SplitRule and Value being folded in as inner
classes.
