Add per-subject bias-corrected intensity histogram QC

[Copilot]

Adds a QC rule that generates a per-subject, per-stain intensity histogram from bias-field-corrected intensities sampled patch-wise from within the brain mask — avoiding the cost of processing the full zarr volume.

Changes

• segmentation.smk / n4_biasfield.py: Add biasfield ome.zarr output to the n4_biasfield rule (mirrors existing gaussian_biasfield behaviour) by passing upsampled_ome_zarr_path to apply_scaled_processing.

• qc.smk — new rule qc_bias_corrected_histogram: Takes the raw SPIM zarr, the downsampled bias field ome.zarr (keyed on correction_method, works for both n4 and gaussian), and the brain mask NIfTI. Parameterised via existing n_patches_per_label, patch_size, and patch_seed config keys.

• qc_bias_corrected_histogram.py — new script:

  1. Wraps the brain mask NIfTI as a single-label ZarrNiiAtlas (no TSV needed) and calls sample_region_patches to draw random patch centres in physical space.
  2. Extracts full-res patches from the raw zarr at level 0; extracts the corresponding region from the bias field at registration_level within its pyramid (covering the same physical extent with patch_size / 2**registration_level voxels).
  3. Upsamples the bias field crop to the raw patch shape via scipy.ndimage.zoom(order=1) (linear; avoids ringing on a smooth field).
  4. Divides raw by bias field per-patch, accumulates intensities, and renders the same 4-panel histogram PNG as qc_intensity_histogram.

• Snakefile: Adds bias-corrected histograms to all_qc for stains_for_seg, gated on do_seg.