Update JudgeRubric Response Logging

[warner-benjamin]

Currently vf-eval doesn't log additional state information. Calling env.evaluate will log custom state info if one passes state_colums: list[str] to make_dataset.

There's an open issue about this on verifiers PrimeIntellect-ai/verifiers#418. We should update our environments to log judge responses when this feature is implemented.

[warner-benjamin]

verifiers now logs the info dict, or at least does with our medarc-eval command, so we want to update all our LLM-as-a-Jusge environments to log the complete judge response to the info dict (or state dict if that works) so we have the Judge responses for reference.