Add support for re-scoring evaluations

[ob1-s]

Description

My attempt to implement #438. I ran into a few edge-cases I need to solve before this can be merged.

I successfully tested the re-score workflow with the following envs: gsm8k, wordle, simpleqa, continuation_quality and reverse_text.

But found some edge-cases in these:

• tool_test: since the re-scoring logic doesn't deserialize the tool call objects from the original results, it fails to apply the rubrics;
• acebench_agent_multistep: if i got this right, the re-scoring doesn't work here because this MultiTurnEnv relies on non-serializable state.

Any direction here would be very much appreciated.

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Test improvement

Testing

• All existing tests pass when running uv run pytest locally.
• New tests have been added to cover the changes
• Relevant tests have been updated.

Checklist

• My code follows the style guidelines of this project as outlined in AGENTS.md
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• Any dependent changes have been merged and published [not relevant]

Additional Notes

[willccbb]

hey -- we have a big refactor in progress which will solve a lot of the open issues, but unfortunately will also break some of your PRs :/ wanted to give a heads up, you can check out the evals branch + my draft PR

[ob1-s]

no worries, thanks for letting me know! i'll check them out