mirror cli args in toml config

[mikasenghaas]

Description

previously, we used EvalConfig to determine the valid fields in CLI but this is (a) confusing UX and (b) leads to some fields being ignored (e.g. all of client_config) because the resolution logic to build the eval expects the CLI syntax. for example:

• if you try api_key_var = "MY_KEY" → rejected by validation as "invalid field"0
• if you try client_config.api_key_var → validation might pass but build_eval_config won't find it

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

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

Additional Notes

---

Note

Bug fix: align TOML validation with CLI args

• Replace valid_fields derived from EvalConfig with an explicit set mirroring CLI flags (env, model/client, sampling, evaluation, logging, saving) in load_toml_config
• Global and per-[[eval]] validation now checks against these CLI-aligned fields, reducing false rejections and ignored fields during config resolution

^{Written by Cursor Bugbot for commit 200e08f. This will update automatically on new commits. Configure here.}