feat(plotting): add ncols to wrap multi-panel violin plots (#1552)

[Hore01]

Summary

Adds an ncols parameter to sc.pl.violin so multi-key violins can wrap into a grid instead of one giant horizontal row. Mirrors the ncols convention already used by sc.pl.umap / sc.pl.embedding.

Closes #1552.

What changes

Two code paths in violin() get the new parameter:

• multi_panel=True (seaborn catplot path), forwards ncols to seaborn's col_wrap=. The existing stripplot and rotation loops used g.axes[0] 2D indexing, which col_wrap flattens, so I swapped them to g.axes_dict, same semantics either way, but compatible with both shapes.
• groupby set, multiple keys (legacy setup_axes path) — when ncols is provided, builds a gridspec figure via _panel_grid (already used by sc.pl.embedding). The import is local to the function to dodge the _tools/__init__.py ↔ _anndata.py circular-import (which I hit on the first try — the package's _tools/__init__.py already imports ranking
  from _anndata.py, so a module-level import would cycle).

ncols=None (default) flows through the unchanged code, so every existing test_violin image baseline stays byte-identical.

Why hard-code hspace/wspace

embedding exposes both; doing the same here would add API surface for no gain on the issue as filed. Defaults (0.25, 0.1) match embedding. Easy follow-up if anyone asks.

Tests

Added one new subtest in test_violin:

with subtests.test("ncols"):
    sc.pl.violin(pbmc, ["n_genes", "percent_mito", "n_counts"],
                 multi_panel=True, ncols=2, show=False)
    assert len(plt.gcf().axes) == 3
    save_and_compare_images("violin_ncols_wrap")

The assert len(...) == 3 is cheap insurance; tol=40 would happily mask an off-by-one in panel count, and an earlier draft of this patch had exactly that bug (the original Path B only made one axis when groupby is None because the data is melted into a single column, which is why I ended up wiring the wrap into multi_panel=True instead). The baseline at
tests/_images/violin_ncols_wrap/expected.png was generated locally and visually checked.

Full test_violin (all four subtests) passes locally.

Refs

• Issue requesting ncols parameter in sc.pl.violin() #1552 (open since 2020, good first issue)
• Pattern: _panel_grid in src/scanpy/plotting/_tools/scatterplots.py
• Convention: docs/release-notes/<PR>.feat.md; placeholder added, will rename to the real number in a follow-up commit

[codecov]

Codecov Report

❌ Patch coverage is 91.66667% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 79.64%. Comparing base (2ae768e) to head (7e24234).
⚠️ Report is 2 commits behind head on main.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/scanpy/plotting/_anndata.py	91.66%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4132      +/-   ##
==========================================
+ Coverage   79.61%   79.64%   +0.02%     
==========================================
  Files         120      120              
  Lines       12786    12794       +8     
==========================================
+ Hits        10180    10190      +10     
+ Misses       2606     2604       -2     

Flag	Coverage Δ
hatch-test.low-vers	78.89% <91.66%> (+0.05%)	⬆️
hatch-test.pre	79.50% <91.66%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/scanpy/plotting/_anndata.py	85.65% <91.66%> (+0.22%)	⬆️

... and 1 file with indirect coverage changes

[Hore01]

Rebased on main — the earlier test (low-vers) failure was an unrelated test_logfile flake (passing on the fresh CI run). Also added two subtests covering the new ax/ncols branches so codecov/patch is green.

All CI checks pass except Check title, milestone, and labels, which I can't clear myself as an external contributor. Could you attach the appropriate milestone (or the no milestone label) so it goes green? Happy to address any review comments.