Add nfolds argument to Stacked Ensemble to enable cross-validation

Just like all the other H2O algos, we should be able to perform k-fold cross-validation of the ensemble. To do a full cross-validation of the base model training + ensemble process, its a very computationally heavy operation. I am looking into the idea of just cross-validating the metalearner instead and reporting those metrics/fits. This is actually the only thing that makes sense to support anyway through our current API which expects to be given the already-cross-validated base models (containing cv preds).