feat(evaluators): new api from runners-api

[namrataghadi-galileo]

Summary

• Refactored Galileo Luna invocation so Agent Control calls runners-api directly at /api/v1/scorers/invoke.
• Switched Luna auth to internal JWT only, signed with GALILEO_API_SECRET_KEY or GALILEO_API_SECRET.
• Made scorer_id the required runtime identity; scorer_label is optional metadata and scorer_version_id is optional pinning.
• Ensured requests always include config: {} and never forward Galileo-API-Key to runners-api.

Scope

• User-facing/API changes:

  • Luna configs now require scorer_id; label-only and version-only configs are rejected.
  • Galileo Luna examples/docs now document GALILEO_RUNNERS_API_URL, GALILEO_LUNA_SCORER_ID, and internal secret auth.

• Internal changes:

  • Replaced API /scorers/invoke and /internal/scorers/invoke client routing with runners-api routing.
  • Removed Luna auth mode selection and API URL/console URL derivation.
  • Added tests for endpoint, JWT payload, scorer ID validation, version forwarding, API-key stripping, and request body shape.

• Out of scope:

  • Agent Control UI changes.
  • runners-api/API service implementation changes.
  • Scorer hydration, version resolution, caching, retries, or protect execution logic inside Agent Control.

Risk and Rollout

• Risk level: medium
• Rollback plan:
  • Revert this PR to restore the previous API-mediated Luna invocation path.
  • If rollout exposes saved controls without scorer_id, migrate those controls or temporarily roll back before retrying cutover.

Testing

• Added or updated automated tests
• Ran make check (targeted package validation was run instead)
• Manually verified behavior

Targeted validation:

• make -C evaluators/contrib/galileo test -> 86 passed
• make -C evaluators/contrib/galileo lint -> passed
• make -C evaluators/contrib/galileo typecheck -> passed
• git diff --check -> clean

Checklist

• Linked issue/spec: RFC on direct runners-api scorer invoke
• Updated docs/examples for user-facing changes
• Included required follow-up tasks: confirm/migrate any legacy saved Luna controls missing scorer_id

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!