refactor(config): remove the unread ocr_det_* region profile fields

[davidamacey]

Closes #181

Stacked on #180 (branch cleanup/dead-symbols-174, which deletes PaddleOcrRegionDetector); that PR's commit shows in this diff until it merges. Merge #180 first.

Changes

• Remove ocr_det_model, ocr_det_version, ocr_det_input_size, ocr_det_prob_floor from DetectionProfile, RegionProfileBody, the license_plate example, docs, tests and contracts (regenerated).
• Remove their remaining readers: range/multiple-of-32 validation, the ocr_det_model Triton readiness check, the unload guard and promoted-models listing entries, the 'OCR text hint' detector vocabulary entry, and the ocr_det_models choices_from value. The ocr.det_models vocabulary list stays.
• ocr_engine_id (stamped as region_text_engine_version) is now <rec model>:<version>.
• A stored or submitted profile still carrying any of the four keys is rejected (profile_field_unknown) with a message naming the key and saying it is retired (RETIRED_PROFILE_FIELDS in profile_registry.py). Tests cover the validate path and the stored-doc load path.
• Frontend: types, fixtures, e2e stub and contract test updated; CHANGELOGs updated.
• OCR_DET_MODEL (settings.py / ocr_service.py) is untouched.

Verification

• Targeted pytest (profile validation, detection profile, region vocabulary/text, config vocabulary, router, unload): 193 passed.
• Frontend: npm run check 0 errors, npm test 4390 passed.
• pre-commit run --from-ref origin/main --to-ref HEAD exit 0.