Add vessels mask skeletonization rule using overlap-aware chunked processing

[Copilot]

This adds a vessels workflow step to generate a skeletonized .ozx mask from the vessel mask, using the same overlap-aware chunking pattern as signed distance transform. The implementation uses dask.map_overlap with scikit-image skeletonization to avoid chunk-boundary artifacts.

• Workflow integration (vessels rule)

  • Added skeletonize_vessels_mask in spimquant/workflow/rules/vessels.smk.
  • Input: datatype=vessels, suffix=mask.ozx, desc={desc}.
  • Output: datatype=vessels, suffix=mask.ozx, desc={desc}+skeleton.
  • Configured overlap depth and resources consistent with existing heavy chunked vessel transforms.

• Skeletonization implementation (new script)

  • Added spimquant/workflow/scripts/skeletonize_vessels_mask.py.
  • Reads vessel mask via ZarrNii, applies per-channel skeletonization in chunked blocks via da.map_overlap, and writes OME-Zarr output.
  • Preserves project mask convention (foreground=100) in output voxels.

• Robustness guardrails

  • Added explicit dimensionality check for expected (c, z, y, x) layout before processing.
  • Documented overlap semantics for channel vs. spatial axes.

skel_darr = da.map_overlap(
    skeletonize_block,
    znimg.darr,
    depth={0: 0, 1: overlap_depth, 2: overlap_depth, 3: overlap_depth},
    boundary=0,
    dtype=np.uint8,
)