Add github action to run lightweight evals as presubmit job

[droot]

We want to run a few eval tasks (may be difficulty level easy) as pre-submit tasks to ensure that new PRs are getting some basic e2e tests.

Details:

• The pre-submit job should bring up a kind cluster
• The pre-submit job should also bring up ollama or llamacpp server
• Run evals with the difficulty level easy or to begin with --task-pattern=scale
• Add contributing section in the README.md to use this job as e2e test

[droot]

/cc @tuannvm let me know if you will be interested in contributing towards this.

[droot]

/cc @sakajunquality

[rxinui]

Hey! I can have a look at that @droot

[droot]

Hey! I can have a look at that @droot

Sure. Go ahead!

[rxinui]

@droot ater analysing the codebase, it seems like the CI is somehow dependant to ./ci/** folder which seems to embed an internal logic I'm not aware of. As a result any attempt to manually update .github/workflows/ci-presubmit.yaml results in an error on step verify-autogen.

Therefore, I create a separate worklow on pull_request.

An analysis is run and display as a GITHUB_SUMMARY_STEP for clarity.

Note: at the moment, tasks are evaluated on the default ollama model.