ID-back scanner guide and decode ROI diverge across orientation

Observed 2026-09-07 from e3-solutions/theft git history (commits ae87effc, 85edec71, 9f935599, 44db6df): the CDL-back rectangle is dynamic only by viewport orientation (92%×18% portrait; 90%×~26.4% landscape) and has content-driven text/status, but its border stays red/dashed for both local and Adam ID-back sessions. Local decoding derives its ROI from intrinsic video orientation (portrait ~99%×21.96%, shifted 4% left; landscape 96%×~32.22%, shifted 4% left). Adam uploads the full frame and samples a hard-coded centered 92%×18% object-cover-mapped area for guidance, even when the visible landscape guide is 90%×~26.4%. Because CSS uses viewport orientation/object-cover while local uses intrinsic video dimensions, the visible boundary is a positioning aid rather than an exact decode boundary and the two can diverge. Commit 44db6df did not alter ScannerOverlay or these coordinates; it added canary fallback/selection.