Add --bbox_config CLI option for template-space bounding box crops

[Copilot]

Adds a mechanism to extract cropped NIfTI regions from SPIM zarr data in template space at arbitrary resolution, useful for targeted visualization or downstream analysis without registering the full volume.

Changes

CLI & config (snakebids.yml, Snakefile)

• New --bbox_config option accepting a YAML file with a list of named bounding boxes
• Parsed at planning time into bbox_configs / bbox_cfgs_by_name dicts

Rules (patches.smk)

• extract_bbox_crop — per-bbox/stain rule; output BIDS entities: stain, space (template), res (e.g. 20um), desc (bbox name), level
• all_bbox_template_crops — standalone target expanding over all bboxes × dataset stains

Script (extract_bbox_in_template_space.py)

Block-wise pull-resampling via zarrnii interp_by_block. Transform chain:

1. ref ZYX vox → RAS mm (zarrnii ZYX affine)
2. RAS template → RAS subject (composite inverse displacement field)
3. RAS mm → flo ZYX vox (inverse SPIM zarr affine)

Only the subregion of the source zarr needed per output block is loaded — memory stays bounded at any input level.

Bbox config format

bounding_boxes:
  - name: entorhinal
    x: 5.2        # RAS mm centre
    y: -2.3
    z: -1.8
    units: mm
    size: [256, 256, 256]   # [nx, ny, nz] voxels along [R, A, S]
    resolution_um: 20
    input_level: 0
  - name: wholebrain
    use_brain_mask: true    # derives extent from template brain mask
    resolution_um: 100
    input_level: 2

The use_brain_mask: true special case computes the bounding box from the template brain mask non-zero extent, enabling whole-brain resampling at arbitrary resolution.