[Yushu-Li]

Follow up to #519, partially addressing A3 of #468. Add an opt-in config_from_alphaevolve helper alongside the evaluator adapter so local Python examples can reuse their experiment dictionaries with an explicitly selected OpenEvolve endpoint.

Import model names/relative weights, a required generation-attempt budget, and problem/context text. Reject missing budgets/models and unsupported wall-clock limits; warn by field path for unmapped settings such as cloud concurrency. No Google dependency, clients or requests are added. Existing config loading is unchanged; the walkthrough documents the limited mapping and uses the returned Config directly with the Python API.

Validation:

• Full python -m unittest discover -s tests -q: 700 tests passed, including 13 new configuration tests and a real controller/worker run using the imported budget and prompt.
• Black, isort and git diff --check pass.
• An actual AlphaEvolveExperimentConfig.model_dump() from Google's SDK source (674dd5e558e917320c251c085e47fae33e1bbcff) converted successfully and drove one OpenEvolve generation against the unchanged circle-packing seed/evaluator. The provider emitted a fixed test proposal and network connections were blocked. This validates interoperability, not real-model optimization quality; no paid model or Google Cloud calls were made.

Scope: local single-file Python experiment dictionaries only. No .env import, concurrency equivalence, SDK shim, exporter, or strict request/cost budget is claimed. max_programs maps to OE generation attempts, with an additional seed evaluation and possible retries.

AI disclosure: Codex assisted with code, tests and documentation for the user's selected follow-up to the existing #468 issue. All reported validation was executed locally.