Benchmark Foreign Models

[Ceyron]

Adds utility (and a guide to come) as to how to benchmark foreign models, e.g., using Flax or PyTorch. Precisely speaking, to benchmark models that are not written in Equinox or disobey the first-order Markovian and array layout format.

Adds new methods to the Base Scenario:

• Return only the ic set for both train, test, coarse train, and coarse test (Related: Should coarse data be warmed up with coarse or reference stepper? #32) together with unit tests
• Perform all metrics based on an externally given rollout

TODO:

• Guide with Flax / PyTorch

Closes #29