Separate crop geometry from codec effects in ROI benchmarks

ROI-crop benchmarks should report pixel-area reduction as the primary crop-size metric and encoded-byte reduction only as a secondary, format-qualified metric: comparing an original JPEG/HEIC to a lossless PNG crop otherwise mixes cropping with codec effects. Also, a Gemini-gated positive-only cohort can measure detector coverage and extraction agreement, but not detector precision/recall or independent extraction accuracy without source labels/manual adjudication. Observed 2026-09-08 while designing COR-4083 from the Shield ROI benchmark contract and prior specialist evaluation artifacts.