[Yushu-Li]

Follow up to #520, partially addressing A3 of #468. Add an opt-in config_from_alphaevolve_env helper so local Python users can reuse model names/weights and a generation budget from an explicitly named AlphaEvolve .env file through the existing dictionary converter.

Support single or numbered models. Require MAX_PROGRAMS_GENERATED; reject conflicting evaluation budgets instead of silently dropping them. Even equal evaluation budgets warn that OE counts generation attempts plus a seed evaluation and possible retries. Cloud/concurrency variables are not imported and warn by name only. Problem text and the OE endpoint/key remain caller-supplied.

Use the optional alphaevolve-env extra (python-dotenv 1.x) with strict parser validation for malformed/duplicate assignments. No environment discovery, interpolation, global environment mutation, script execution, Google dependency or client creation is added. Existing loaders and the CLI are unchanged; document mappings and limitations.

Validation:

• Full python -m unittest discover -s tests -q: 715 tests passed, including 15 new env-import tests. Those 15 also pass with the declared minimum python-dotenv==1.0.0 (full suite used 1.2.4).
• Black, isort and git diff --check pass.
• Imported unchanged upstream circle-packing and TSP example.env files at 674dd5e558e917320c251c085e47fae33e1bbcff; the signal-processing file correctly failed for missing models. A bounded copy of the circle-packing env drove one generation through the real OE controller/worker with unchanged seed/evaluator and a scripted proposal. Network connections were blocked; no paid model or Google Cloud calls were made. This checks interoperability, not real-model optimization quality.

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