feat(evaluators): update Luna client to support additional metrics

[namrataghadi-galileo]

Summary

• Added structured runtime context for Galileo Luna scorer invocation while preserving backward compatibility.
• Extended Step with optional tools and ground_truth fields.
• Added a contextual evaluator hook that receives selector-selected data and the complete runtime Step.
• Updated galileo.luna to dual-write legacy inputs and the expanded structured record.
• Added reliable available-tool capture for Google ADK and Strands.

Scope

• User-facing/API changes:
  • Added optional tools and ground_truth parameters to the Python SDK’s evaluate_controls.
  • Added optional tools and ground_truth fields to the generic OSS Step model.
  • Existing SDK calls and Step payloads remain valid without modification.
  • Existing evaluators implementing only evaluate(data) remain compatible.
• Internal changes:
  • Added evaluate_with_context(data, step) with a default delegate to evaluate(data).
  • Updated the engine to pass the complete runtime Step without changing selector behavior.
  • Updated Luna requests to send legacy inputs.query, inputs.response, inputs.ground_truth, and inputs.tools alongside the structured record.
  • Mapped Step.ground_truth to record.dataset_output.
  • Added normalized tool-definition extraction for Google ADK and Strands.
  • Preserved scorer ID/version/configuration, local threshold/operator evaluation, error metadata, and existing response parsing.
• Out of scope:
  • Tenant authentication or authorization changes.
  • Project, organization, target, or log-stream resolution.
  • Runners implementation changes.
  • Wizard, Protect, or obsolete API scorer-invoke integration.
  • galileo.llm and galileo.code evaluators.
  • Removal of the legacy query/response representation.

Risk and Rollout

• Risk level: medium
• Rollback plan:
  • Revert the AgentControl change and continue using the legacy Runners inputs.query/inputs.response contract.
  • Keep the Runners legacy adapter enabled throughout rollout.
  • Do not enable the new preset runtime path until the additive Runners contract is deployed.
  • Keep tenant-created Luna scorers disabled until trusted authorization support is available.
  • Retain dual-write until older supported Runners versions have been retired.

Testing

• Added or updated automated tests
• Ran make check using UV_OFFLINE=1 make check
• Manually verified legacy and structured Luna request payloads through HTTP contract tests

Checklist

• Linked issue/spec (if applicable)
• Updated docs/examples for user-facing changes
• Included any required follow-up tasks
  • Deploy the additive Runners contract before enabling the structured preset runtime path.
  • Enable tenant-created Luna scorers only after the separate authentication work is complete.
  • Remove legacy dual-write in a future change after older Runners versions leave the supported deployment window.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!