Genepro

fix bug in calling subtree mutation

- Now score, internal_nodes, and leaf_nodes are handled different in the scikit-learn estimator.
- Better default hyperparam settings are set for genepro
- More nodes have been implemented

now one can separately call "initialize_population" and "perform_generation", which allows performing intermediate computations if desired

- fix random donor was assigned to entire pop rather than per crossover event
- implemented alternative crossover operator
- clean up some unimportant lines in gym example

Now a Node object n has:
- len(n) -> returns the number of nodes in the subtree rooted at n
- n(X) -> shorthand for n.get_output(X)