Replace fixed histogram range/bins in multiotsu with percentile-based range and bin-width; add threshold sweep QC report

[Copilot]

The multiotsu histogram used a hardcoded intensity range [0, 1000] with 1000 bins, which breaks on datasets with different dynamic ranges. This replaces those fixed parameters with a percentile-based range and bin-width approach, and adds an optional target rule to visually optimize the threshold.

Parameter changes

• --seg_hist_range → --seg_hist_percentile_range (default: [1, 99], type: float) — range adapts to each dataset
• --seg_hist_bins → --seg_hist_bin_width (default: 1, type: float) — bin count derived as ceil((range_hi - range_lo) / bin_width)

multiotsu.py updates

Loads a downsampled pyramid level (tries levels 5→1) to compute percentiles efficiently, then uses the resulting range and bin width to build the histogram before running multi-Otsu thresholding.

New: all_otsu_hist_qc target rule

Invoke to produce one self-contained HTML report per subject/stain/method:

snakemake all_otsu_hist_qc

Each report contains:

• The multi-Otsu histogram PNG (from the multiotsu rule)
• A grid of 2D axial crops with the binary mask overlaid in red, across 10 evenly-spaced threshold values spanning the 1st–99th percentile range

This allows visual inspection to pick the best threshold before committing to a full segmentation run. The rule reuses the bias-field corrected zarr already produced for the multiotsu rule (temp file kept alive until both consumers finish).