Get Eval KPI

Fetch the score trend for one eval as a time series over its completed runs: mean submitted-side score per day/week/month, the improved-side ceiling, refusal rate, and the current-vs-prior-window delta. Read when reporting whether writing quality is improving over time. The trend tracks the grade of what callers SUBMITTED, read from verdict.headline so it means one thing across an eval_version bump; windows under 3 scored runs abstain with a named reason rather than assert a mean.