Fix Wald thresholds for multi-coefficient and reversed contrasts

[dnncha]

Wald thresholds and directional truncation currently operate on each fitted coefficient before applying the contrast. For a multi-coefficient or reversed contrast, this tests a different hypothesis from the scalar effect contrast @ lfc and can make significance depend on the model reference level.

Apply the null and all four alternative-hypothesis transformations to the scalar contrast effect, retaining the existing contrast standard error. The ordinary nonzero-null calculation is corrected in the same way.

A public-API reproduction with fixed synthetic negative-binomial counts (90 samples, three groups, 100 genes) compares C versus B with lfc_null=1.5 and alt_hypothesis="greaterAbs". The estimated log2 fold change is approximately 0.986 with SE 0.0366 in both encodings. Before this patch, reference A produces p=1.53e-159 while reference B produces p=1; after it, both produce p=1. This reproduces in released 0.5.4 and the affected logic is present at current base 95ace12cfa39b6d5982c1867d8127a6b5e9e5134.

Validation: 40 added scalar-normal-reference cases cover all five test modes, zero/nonzero nulls, multi-coefficient contrasts, reversed contrasts and positive singleton controls. Before the patch, 20 fail and 20 pass; all 40 pass after. Full local suite: 105 passed on Python 3.12. Ruff passes for the changed files. Existing R-derived reference fixtures pass; no live R/DESeq2 execution is claimed. This does not establish an effect on a published analysis.

Prepared with AI assistance. Audit and downloadable reproduction inputs, patches and execution records.

The upstream pre-commit.ci formatting-only commit was pulled locally; all 40 focused checks pass on final head 68f16c538f4e93ce3f2880aa4e713a330496a898. The full-suite result above is from authored commit a779e7bee6dc6791c1ff9666e781e4ec83e03c62.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.90%. Comparing base (95ace12) to head (68f16c5).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #468      +/-   ##
==========================================
+ Coverage   85.89%   85.90%   +0.01%     
==========================================
  Files          15       15              
  Lines        1276     1277       +1     
==========================================
+ Hits         1096     1097       +1     
  Misses        180      180              

Files with missing lines	Coverage Δ
src/pydeseq2/glm.py	81.52% <100.00%> (+0.20%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.