feat: switch scan engine to ML Kit Document Scanner (A plan, v1.003)

Custom CV pipeline (CameraX + OpenCV detection + manual editor) could
not robustly handle glare/perspective. Replace the scan step with the
Google ML Kit Document Scanner (on-device AI: live edge detection,
perspective, filters, capture guidance). Keep the custom shell — home,
CamGallery, OCR, share image/PDF choice, version display. Remove CameraX,
OpenCV, custom camera/editor/detector/filters and the CAMERA permission.

Constraint: scan UI is Google-owned; runtime permissions back to zero; OCR retained.
Confidence: high
Scope-risk: module
Reversibility: moderate (git history retains custom pipeline)
Directive: user trials A; B (custom TFLite) deferred.
Tested: build + unit tests (OcrTextJoiner, MediaSaver) pass; scanner device-verified by user.
Not-tested: on-device ML Kit scan flow.
Related: 2026-05-31-clearcam-mlkit-scan-engine-design.md

🐙 Autopus <noreply@autopus.co>