Add evaluation result insights

[fengting124]

Summary

• add a tested evaluation insight utility for sample outcome counts, confusion matrix data, and label distribution
• render Sample Outcome, Label Distribution, and Truth By Prediction panels on /admin/evaluations
• document the design, implementation plan, worklog, and deferred scope
• align the evaluation workbench shell panels after screenshot review

Why

The evaluation page had run status and aggregate metrics, but it did not explain where the model was correct or confused. This keeps the current UI style and avoids backend/API expansion while making model-quality discussion clearer for interviews.

Verification

• npm run test
• npm run lint
• npm run build
• mvn -B test
• git diff --check
• Chrome headless screenshot against a production build with mocked evaluation API data

Notes

• No model weights were downloaded.
• No backend aggregation API was added.