Refactor tests for benchmark metrics (#3183)

Summary:

**Context**: Currently, benchmark metric tests conflate whether the metadata consumed by `fetch_trial_data` has multiple time steps with whether the output data should be `MapData`, and don't explicitly address the separate cases of whether or not there is a `BackendSimulator` attached to the metadata. They are also not organized well. These issues will become more pressing when we start to allow `BenchmarkMetric` to consume data with multiple time steps and/or use a backend simulator

**This diff**:
* Organizes those unit tests better
* Adds some more checks

Reviewed By: Balandat

Differential Revision: D67281379