Honor the requested coefficient in shrinkage grid fallback

[Intron7]

When a two-coefficient shrinkage fit does not converge, nbinomGLM falls back to a grid that always applies the Cauchy prior to coefficient 1. A request to shrink coefficient 0 therefore changes its objective during recovery.

Pass the requested coefficient through the fallback and into nbinomFn. The new shrink_index=1 argument is appended to grid_fit_shrink_beta, preserving existing positional calls and default behavior.

The regression forces optimizer failure, compares the grid result with an independently optimized negative-binomial log-PMF plus priors, and verifies that permuting design columns while adjusting the requested index preserves the fit. Both parameterized cases fail before the fix and pass afterward.

Validation: full upstream suite passes (67 tests); Ruff lint, formatting, and mypy pass.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.53%. Comparing base (95ace12) to head (9f5147a).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #472      +/-   ##
==========================================
+ Coverage   86.05%   87.53%   +1.48%     
==========================================
  Files          15       15              
  Lines        1276     1276              
==========================================
+ Hits         1098     1117      +19     
+ Misses        178      159      -19     

Files with missing lines	Coverage Δ
src/pydeseq2/distributions.py	100.00% <ø> (+1.81%)	⬆️
src/pydeseq2/grid_search.py	70.00% <ø> (+25.71%)	⬆️

🚀 New features to boost your workflow:

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