User story: optimal handling of multiple model runs

This user story articulates a long-term goal for how CMEW should handle multiple model runs.

Long-term goal:
- As a model developer who wants to understand the performance of multiple test runs,
- when I provide CMEW with a list of N model development and CMIP runids, one of which is identified as a reference run,
- and when I provide a list of ESMValTool recipes,
- Each CMEW recipe should support an arbitrary number of recipes
- (For plots / statistics that support only 2 datasets, the recipe should perform these N-1 times. Each should compare the reference run against one of the other runs).
- If the recipe supports the concept of a reference simulation, it should use the indicator provided by CMEW. (If it doesn't support the concept of a reference, it will just ignore this.)