Add --use-work-zarr option for intermediate OME-Zarr rechunking before segmentation/vessel processing

[Copilot]

When input SPIM data has suboptimal chunking or lives on slow network/cloud storage, bias-field correction and vessel segmentation rules suffer from inefficient random reads. This adds an optional pre-processing step that rechunks the full multi-channel, multi-scale zarr to a fast local disk before those rules run.

New option

--use-work-zarr: requires --work_dir to point to a fast local disk with sufficient capacity. Raises ValueError at workflow init if --work_dir is unset.

spimquant /bids /output participant \
  --work_dir /fast/local/nvme \
  --use-work-zarr

Changes

snakebids.yml

• New --use-work-zarr flag with help text noting the --work_dir requirement

Snakefile

• Validates --work_dir is set when --use-work-zarr is active
• Defines work_zarr_path (always .ome.zarr directory, never .ozx, rooted at work_dir)
• Defines spim_path_for_seg_vessels — routes downstream rules to either the rechunked work zarr or the original SPIM path
• Pre-populates spim_json_overrides for rechunked paths so JSON sidecar channel/orientation overrides remain effective
• Includes new rules/rechunk.smk

rules/rechunk.smk (new)

• rechunk_spim_to_work rule: one output per subject, contains all channels and all scale levels; output is directory(work_zarr_path)

scripts/rechunk_spim.py (new)

• Traverses the zarr group tree recursively with dask rechunking; preserves all .zattrs metadata verbatim
• Supports .ome.zarr directories, .ome.zarr.zip, and .ozx inputs; raises a clear error for .ims
• Writes zarr format 2 (OME-NGFF v0.4/v0.5 compatible with ZarrNii)
• Closes zip stores explicitly in a finally block

rules/segmentation.smk, rules/vessels.smk

• gaussian_biasfield, n4_biasfield, and run_vesselfm input spim now references spim_path_for_seg_vessels instead of inputs["spim"].path directly