fix: shift signed CLR only for differential_abundance Wilcoxon

[elhb]

Description

Makes differential_abundance valid on signed CLR.

difference is still mean(target) - mean(reference) on the original
matrix, matching PixelatorR RunDAA (fc.name = "difference"). Wilcoxon
runs on a test-only copy where each marker is shifted so its minimum is 0.
That shift is the same for every cell, so ranks (and thus p-values) match a
Wilcoxon on the original CLR. Scanpy still computes log-fold changes
internally, which break on negatives; we do not use those values.

Based on PNA-3531.

Fixes: PNA-3532

Type of change

• Bug fix (non-breaking change which fixes an issue)
• This change requires a documentation update

How Has This Been Tested?

uv run pytest tests/pna/analysis/test_differential_abundance.py

• test_differential_abundance_returns_expected_columns
• test_differential_abundance_layer_from_obsm
• test_differential_abundance_group_vars_splits_rows
• test_differential_abundance_invalid_reference_raises
• test_differential_abundance_signed_clr_keeps_original_difference_sign

PR checklist:

• This comment contains a description of changes (with reason).
• I have performed a self-review of my own code
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• If a new tool or package is included, I have updated dependencies in pyproject.toml and cited it properly
• I have checked my code and documentation and corrected any misspellings
• I have documented any significant changes to the code in CHANGELOG.md

---

Note

Low Risk
Scoped change to PNA differential abundance statistics; non-negative CLR paths are unchanged, with clearer behavior for signed CLR effect sizes and p-values.

Overview
differential_abundance now supports matrices with negative values (e.g. signed CLR) without relying on scanpy Wilcoxon on raw signed data.

When negatives are present, the helper builds a test-only layer (_pixelator_daa_nonneg) by shifting each marker so its column minimum is at least zero (same shift for every cell, preserving ranks). scanpy.tl.rank_genes_groups Wilcoxon p-values use that layer; difference, pct_1, and pct_2 stay on the original selected matrix, aligned with pixelatorR RunDAA. The previous runtime warning filter and negative-value warning are replaced by a one-time info log. Docs and CHANGELOG describe the new flow; the signed-CLR test now asserts the reported mean difference matches the original matrix and that input adata.X is unchanged.

^{Reviewed by Cursor Bugbot for commit 3a3b424. Bugbot is set up for automated code reviews on this repo. Configure here.}