Add the functionality to automatically register images in the "Camera Registration" tab

[romleiaj]

Uses MINIMA-LoFTR under the hood to compute features and match points - a model tuned to work across modalities (EO/IR)

[romleiaj]

See VIAME/VIAME#272

[romleiaj]

Branch summary: dev/auto-image-registration

What it adds: an "Auto Register" button in the Camera Registration panel that computes a cross-modality (e.g. EO↔IR) homography between two cameras automatically, using a deep feature matcher (MINIMA-LoFTR) hosted in the VIAME desktop interactive service. ~550 added lines over 14 commits.

How it workds

• Interactive service / IPC (interactive.ts, ipcService.ts): a new register_images wire command sends two image paths (plus optional RANSAC/matching thresholds) to the Python service, which returns a 3×3 homography, the inlier correspondences, and quality stats. The matcher model loads lazily on first use and stays resident. A companion register-images-available handler probes whether the matcher weights exist in the VIAME install; without them the button is hidden entirely. The boundary is typed end to end: both the IPC handler and the service method use the shared AutoRegisterRequest/AutoRegisterResponse types, with a local interface for the service's snake_case wire payload.
• API contract (apispec.ts): new AutoRegisterRequest/AutoRegisterResponse types, including machine-readable failure codes (insufficient_matches, low_confidence, degenerate_homography).
• Platform bridge (useAutoRegister.ts, Viewer.vue, ViewerLoader.vue): a provide/inject service keeps the shared UI platform-agnostic — desktop's ViewerLoader passes a handler prop that resolves each camera's image path for the current frame and calls the service; on web nothing is provided, so the panel shows no button. Along the way, the stereo-only image-path-getter setup in ViewerLoader was refactored into a shared populateMultiCamImagePathGetters, because Auto Register works on any multi-camera dataset, not just stereoscopic ones (this does touch stereo code).
• UI + store (RegistrationTools.vue, CameraRegistrationStore.ts): the button runs the matcher on the current frame's pair, confirms before replacing any existing points/transform, then injects the returned inliers as ordinary correspondence points — so the user can inspect, drag, or delete them exactly like hand-picked ones — switches the pair to a homography fit, and refits. Point editing is auto-enabled so the injected points are immediately visible. Results surface as a success summary (point count + match consensus) or an actionable error.
• Copy changes: the feature was renamed Auto Align → Auto Register (UI, IPC, and wire names all updated), and the "Pick points" toggle became "Edit points." The button reads "Auto Register" with the current-frame qualifier in a tooltip. The rig status now says "N/M cameras ready" instead of "registered" (the count includes the reference camera, which is ready by definition but was never registered to anything) — this touches the panel summary, the Align View tooltip, and the import menu hint.

Tests

CameraRegistrationStore.spec.ts gains coverage for applyAutoRegistration (point injection, homography switch, source non-restamping), and alignedViewTooltip.spec.ts is updated for the "cameras ready" wording. The service/IPC layer and Vue wiring are untested — the Python matcher side lives in the separate VIAME/service repo, not this diff.