Add Advanced Edge Detection + active-mask layer

[jacobson30-bot]

Summary

Adds Canny and Sobel/Scharr edge detection as a non-destructive tool whose outputs become reusable analysis objects, plus a formal active-mask layer (ImageMask + MaskSet) that parallels the existing ROISet.

Workflow: select method → tune → live overlay preview (never mutates the image) → emit overlay / new image / active mask / ROI(s) → mask restricts statistics, excludes regions from a plane fit, converts to ROI.

What's included

• Edge backend — processing/edge_detection.py (EdgeDetectionResult, canny_edges(), gradient_filter(), CANNY_PRESETS; NaN-safe, ROI-aware, provenance-recording). edge_detect extended with sobel/scharr (history-replayable). scikit-image promoted to a core dependency.
• Active mask layer — core/mask.py (ImageMask + MaskSet, packbits-compact JSON), processing/mask_ops.py (9 morphology ops), core.roi.roi_from_mask() on the existing Shapely bridge, {stem}.masks.json sidecars mirroring the ROI sidecar.
• GUI — modeless EdgeDetectionDialog with live overlay preview + Process-tab button; canvas set_mask_overlay/clear_mask_overlay; viewer mask mixin with sidecar autosave; MaskManagerPanel + "Masks" sidebar tab (set-active, cleanup, invert, convert-to-ROI, statistics, export).
• Downstream — active mask restricts statistics (roi_statistics(mask=…)) and excludes regions from a plane fit via the mask→ROI bridge.

Verification

• 1961 passed, 207 skipped, ruff check clean across the package.
• New backend tests (62 passing): edge detection, mask model/ops/sidecar/integration.
• Offscreen-Qt dialog + manager tests written and gated; they skip where Qt has no platform plugin and run in CI.

Note: the live Qt GUI path was verified statically (imports, wiring, construction order) but not executed in the dev sandbox (no Qt platform plugin). Deferred clean seams left for Gabor/ridge detectors and edge-orientation histograms.

🤖 Generated with Claude Code