Add AI-ready image and vision foundation

[rsasaki0109]

What changed

• add typed image ownership, safe mutable/strided ROI views, planar layouts, and color metadata
• add feature-gated spatialrust-vision preprocessing, warps, detection post-processing, masks/RLE, dense maps, and spatial bridges
• add camera/RGB-D integration, Python bindings and stubs, OpenCV comparisons, benchmarks, property tests, and end-to-end demos

Why

SpatialRust needs a Rust-native, AI-ready 2D image foundation that connects calibrated camera data and learned dense outputs to the existing point-cloud pipeline without hidden device transfers or mandatory heavy dependencies.

Validation

• cargo test --workspace
• cargo test -p spatialrust-vision --features full
• cargo test --workspace --all-features --no-run
• Python: 48 pytest cases plus mypy.stubtest
• OpenCV 4.13 numerical comparison: pass
• all vision-* features checked independently
• Criterion preprocessing benchmark and executable vision-to-MVP demo