test(agent): characterize configuration compatibility

[zpzjzj]

Summary

• document the current engine, protocol, provider, model, auth, and local-login semantics as a compatibility contract
• add characterization tests for the public GitHub Action input-to-CLI/environment translation
• keep historical --model provider/name, opaque slashed model IDs, and empty-engine delegation documented
• run the Action adapter tests in GitHub CI with an explicitly provisioned Python version

Scope

This is phase 1 of #196. It intentionally does not change runtime configuration behavior or add --provider; it establishes the compatibility boundary for the later resolver and adapter refactors.

Verification

• make verify
• make test
• make test-action

Part of #196

[JHWang-1997]

The tests cover argv construction and engine_env() separately, but not the final composition of engine_env(), engine_model_env(), and provider_env() in main(). A future refactor could break Agent Judge model selection or Codex custom-provider routing while all tests still pass. Please add a characterization test for the complete argv/environment assembly path.

[zpzjzj]

Addressed in 0e8752f. Added test_main_composes_codex_argv_and_environment, which exercises the complete main() orchestration path with external operations mocked, then asserts the final argv, Agent Judge OPENAI_MODEL, engine auth/base URL, and Codex custom-provider environment. Verified with make verify, make test, and make test-action.

[JHWang-1997]

LGTM