Fix PyDESeq2 layer handling in fit()

[giovp]

Problem

PyDESeq2(..., layer=...) validates the layer at init, but .fit() passed the original AnnData to PyDESeq2, which always reads adata.X. If X is non-integer (e.g. normalized), .fit() fails with ValueError: The count matrix should only contain integers.

Fix

When layer is set, build an internal counts-only AnnData with .X = adata.layers[layer] (no mutation of user .X) and pass that to DeseqDataSet.

Test

Add regression test where .X is non-integer but layer contains integer counts; assert .fit() succeeds and .X is unchanged.

[giovp]

good agent

I would have described it differently but good job for just clicking the mouse 😅

[codecov-commenter]

Codecov Report

❌ Patch coverage is 75.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 71.91%. Comparing base (12897e1) to head (d767b39).
⚠️ Report is 19 commits behind head on main.

Files with missing lines	Patch %	Lines
...y/tools/_differential_gene_expression/_pydeseq2.py	75.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #895      +/-   ##
==========================================
- Coverage   73.54%   71.91%   -1.63%     
==========================================
  Files          48       48              
  Lines        5613     5736     +123     
==========================================
- Hits         4128     4125       -3     
- Misses       1485     1611     +126     

Files with missing lines	Coverage Δ
...y/tools/_differential_gene_expression/_pydeseq2.py	92.50% <75.00%> (+0.60%)	⬆️

... and 8 files with indirect coverage changes

🚀 New features to boost your workflow:

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

[Zethson]

Thank you very much!

I generally feel like layer support should rather be improved upstream in pydeseq2 but if you need this now, I'm happy to merge this PR.

[giovp]

Addressed all maintainer feedback:

• Removed the unnecessary test_adata.copy() in the new regression test (fixture is already per-test).
• Moved numpy import to module scope (no local import in the test).

Thanks!

[giovp]

it's more that if pertpy accepts layer as argument, then I expect that that would be passed to the underlying pydeseq dataset.