ci: run pytest on push and PR

[b-erdem]

What this does

Follow-up to #18. Adds a GitHub Actions workflow that runs the test suite on push to main and on pull requests. Until now the only workflow was docker-publish.yml, so nothing ran the tests automatically.

The job installs the package with dev extras, sets up Temurin JDK 17, and fetches tla2tools.jar (pinned to the same v1.8.0 that scripts/setup_tools.py uses), so the TLC-gated behavioral tests in tests/test_languages actually run in CI instead of skipping.

Scope

Runs tests/test_languages. tests/test_models is intentionally left out for now:

• test_model_connection.py makes live provider API calls (reads OPENAI_API_KEY, ANTHROPIC_API_KEY, etc.), so it needs secrets and a network and is not safe to run in CI.
• test_litellm_adapter.py has two failures that already exist on main and are unrelated to this workflow (test_resolve_model_name_for_custom_openai_compatible_api and test_model_factory_reports_litellm_as_recommended_provider). They look like the model adapter and its tests drifted apart. Happy to send a separate PR to sort those out, after which this workflow can be widened to the whole tests/ tree.

Notes

• tla2tools.jar is fetched at the pinned version rather than via sysmobench-setup, to keep the CI step independent of the broader environment checks in that script.
• Python is pinned to 3.11. Easy to turn into a matrix later if you want to cover the requires-python >=3.8 range.

[b-erdem]

@Qian-Cheng-nju heads up from setting this up: two tests in tests/test_models/test_litellm_adapter.py fail on current main, unrelated to this PR, which is why I scoped CI to tests/test_languages here.

• test_model_factory_reports_litellm_as_recommended_provider raises KeyError: 'legacy_providers' (the dict from list_available_models() has no such key).
• test_resolve_model_name_for_custom_openai_compatible_api expects openai/deepseek-reasoner but gets deepseek/deepseek-reasoner (saw this on Python 3.14, might be litellm-version sensitive).

Are these known or expected? If not I'm happy to open an issue with a repro, or send a fix once we know whether the tests are stale or the behavior changed. Then the workflow here can widen to the whole tests/ tree.

[Qian-Cheng-nju]

Thanks for flagging! @b-erdem Both are just stale tests, already fixed and merged in #20.
I think we can add tests/test_models/test_litellm_adapter.py to the workflow. We can keep test_model_connection.py excluded, it makes live provider calls that need secrets/network, so CI can cover those two files but not the whole tests/ tree.