Decouple Ensemble Step from ML Step

[ntalluri]

With the updates to the code in #160, algorithms with multiple parameter combinations will now run aggregate per-algorithm ML code. As a result, we won't be able to run ensemble code for algorithms with one or fewer parameter combinations. To ensure we have ensemble networks for all algorithms, we need to create a new rule that uses the ensembling code per algorithm, rather than running all the ML code for each algorithm.

[ntalluri]

For later iterations of the evaluation code, having the decoupled ensemble step will be beneficial for implementing the ensemble evaluation idea.

[agitter]

Closed by #175