Evaluate ensemble forecasts

Extend evalml to also work for ensemble forecasts.

### Changes required
* realization as an additional dimension in forecasts and baselines
* additional metrics for ensemble forecasts
* configurable or automatically determined set of metrics and scores for deterministic / ensemble forecasts

Start by reviewing evalml's existing deterministic forecast evaluation pipeline, including how forecasts, baselines, metrics, and scores are represented. Identify the entry points and tests covering those components before deciding how realization dimensions and ensemble-specific metrics should fit. Done means ensemble forecasts are evaluated with an appropriate, configurable or automatically selected metric set alongside deterministic forecasts.

Written by the indexing model from the issue text.