feat(AIC-3449): Support pulling in configs for evals from code

[aknight-ld]

Current build of evaluations from the SDK requires the user to define a generation dict by hand. Callers who want to evaluate a variation that already exists have to copy its model, prompt, and tools by hand. This PR provides functionality to let them name ai_config + variation, fetch that variation's definition, and treat any explicit argument as an override.

---

Note

Overview
Evaluations run() can seed settings from an existing AI Config variation instead of requiring a fully hand-built generation dict. Callers pass ai_config and variation; the SDK fetches the latest variation (and linked model config) via the management API, builds defaults for model/provider/parameters, prompt, tools, and judges, then applies optional generation, tools, and criteria as overrides.

generation is now optional when both config keys are provided. _merge_generation layers overrides field-by-field (parameters merge by key; supplying instructions or messages replaces both fetched prompt slots). Validation is split into config-source checks (before any I/O) and post-merge generation checks. Variation-attached tools require implementations or the run fails early; attached judges become default criteria when criteria is omitted. Warnings are logged when a variation pins an older tool version than the API’s latest.

Adds AIConfigVariation parsing and runner helpers _fetch_config_variation / _fetch_model_config, plus broad tests for seeding, overrides, validation, and edge cases.

^{Reviewed by Cursor Bugbot for commit 510d21d. Bugbot is set up for automated code reviews on this repo. Configure here.}

[donei003]

Approved. But had some feedback which we can incorporate

[donei003]

What would happen if this wasn't true?

[aknight-ld]

this check represents validation to make sure that if a model config is set, it's set correctly. I'm restructuring it to be clearer and adding an exception to fail loudly

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 39a7e2e. Configure here.}