Get Eval Results Trend

Eval score trend from the evalresult fact table — one series per (source step, rubric), one point per scored execution, across EVERY path that scores (workspace runs, API runs with run_evals, batch runs, and later policy sampling).

pl_-prefixed targets chart the pipeline's executions; anything else charts the custom processor's. Both hit their composite index ((team, pipeline|processor, created)). Capped at 5000 rows per window; total_points reports the pre-group count so callers can see truncation.

Cached in Redis DB 3 for 5 minutes per (team, target, window).