calibrate a confidence threshold from one inference pass

Threshold calibration ran the model over every tile once per threshold — three
GPU passes to compare 0.50, 0.25 and 0.15 on a hundred-tile raster. The answer
is already in a single run at the lowest value: detections above a higher cut
are a subset of it, and duplicate suppression walks candidates in descending
confidence, so a lower-confidence box can never displace a higher-confidence
one. The kept set above any cut is identical whichever threshold the run used,
which is what makes one pass sufficient rather than merely cheaper.

QA now takes calibration_thresholds and reads each operating point off the same
precision/recall walk it already performs, marking the F1-optimal cut. The lab
runs inference once and fills its table from the sweep.

The contract test asserted the per-threshold loop by name, pinning the waste it
was meant to describe. It now states what calibration owes an operator: a row
per requested threshold, from one run.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

docs/API_CONTRACTS.md

@@ -1913,6 +1913,18 @@ transaction timestamp, so ordering by `created_at` left the assignment — and
 therefore the score and the false-positive evidence shown to a reviewer —
 undefined between identical runs.
 
+`calibration_thresholds` asks for named confidence cuts alongside the run's own
+operating point. They are read off the same matching pass, so a sweep costs no
+extra inference at all. Each row reports the tally, precision, recall and F1 at
+that cut, and `best_f1_in_sweep` marks the F1-optimal one.
+
+This replaces re-running the model once per threshold. Detections above a
+higher cut are a subset of a lower-cut run, and duplicate suppression walks
+candidates in descending confidence, so a lower-confidence box can never
+displace a higher-confidence one: the kept set above any cut is identical
+whichever threshold the run itself used. Three thresholds therefore cost one
+GPU pass rather than three, and produce the same numbers.
+
 The response also returns `precision_recall_curve`: precision, recall and F1 at
 every confidence value present in the run, plus `average_precision`, `best_f1`
 and `best_f1_threshold`. A single F1 describes one operating point and cannot