perf(vision): accelerate connected components beyond OpenCV

[rsasaki0109]

What changed

• replace queue-based per-pixel connected-component flood fill with row-run scanning and path-compressed union-find
• preserve consecutive row-major labels for both 4- and 8-connectivity while accumulating areas, boxes, and pixel-center centroids analytically
• add a borrowed packed-u8 Rust API and make Python borrow contiguous masks, accept any non-zero foreground byte, and move label storage into NumPy
• add independent flood-fill regressions, non-contiguous Python coverage, Criterion profiles, an OpenCV SAUF harness, GitHub Pages updates, and a dated receipt

Why

The prior flood fill repeatedly queued pixels and generated full neighborhoods inside each component. Structured segmentation and document masks contain long horizontal runs, so labeling runs and joining only overlaps in the previous row removes most queue and neighbor traffic.

Measured impact

On the documented Windows 11 / Intel 6-core host, SpatialRust is 2.17x to 3.61x faster than OpenCV 4.13 CCL_SAUF across structured VGA, 1080p, and 4K segmentation-blob and document-line masks. Labels, areas, and bounding boxes match exactly on all six profiles and 320 additional seeded randomized 4/8-connectivity cases.

The claim is deliberately scoped to structured masks; highly fragmented or dense random noise still favors OpenCV.

Validation

• cargo test -p spatialrust-vision --features dense (19 passed)
• cargo clippy -p spatialrust-vision --all-targets --features dense -- -D warnings
• cargo clippy --manifest-path crates/spatialrust-py/Cargo.toml --all-targets -- -D warnings
• Python binding tests (67 passed, 1 skipped)
• OpenCV report contract tests (8 passed)
• OpenCV harness schema validation and 320 randomized SAUF cases
• changed Rust files pass rustfmt --check
• cargo doc -p spatialrust-vision --features dense --no-deps
• Criterion VGA/1080p/4K structured-mask profiles

Workspace-wide rustfmt still reports the repository's pre-existing formatting baseline in unrelated files; this slice does not modify those files.