fix: anchor sc.pl.scatter colorbar to user-supplied ax

[gkneighb]

Summary

Fixes #3963.

When sc.pl.scatter is called with a user-supplied ax, scatter_base placed the colorbar by fig.add_axes(rectangle) with the rectangle coordinates computed from panel_pos -- positions scanpy uses for its own internal panel layout. Those positions have no relationship to the caller's figure, so the colorbar lands somewhere in the figure interior, typically overlapping a sibling axes.

sc.pl.embedding already handles this correctly by routing through plt.colorbar(sct, ax=ax, ...). This PR detects the user-supplied-ax case in scatter_base and uses the same ax= form. The original rectangle-based path is kept for the scanpy-managed-figure case, so multi-panel scatter outputs that scanpy lays out itself are visually unchanged.

Before / after

The OP's repro pattern (left panel is sc.pl.scatter, right panel is anything else):

Before: colorbar floats between the two panels, partly over the right panel's y-axis labels.
After: colorbar sits directly to the right of the left ax, like every other matplotlib colorbar attached via ax=.

Locally on pbmc3k_processed with fig, axs = plt.subplots(1, 2, figsize=(10, 4)):

• user ax: x0=0.125 x1=0.446
• colorbar: x0=0.449 x1=0.477 (just past the user ax right edge)

Test plan

Adds tests/test_plotting.py::test_scatter_colorbar_uses_user_ax -- structural assertion that the colorbar's x position sits just past the user's ax right edge, rather than overlapping a sibling panel. Uses a tiny synthetic AnnData to keep the test self-contained.

Unable to run the full pytest tests/test_plotting.py locally because every test in that file currently fails for me with Cannot shard v2 format data. Please set anndata.settings.zarr_write_format to 3 from the anndata_settings autouse fixture (tests/conftest.py:162) interacting with anndata 0.12.10 on Apple Silicon. The added test passes when run standalone with explicit ad.settings.zarr_write_format = 3; ad.settings.auto_shard_zarr_v3 = False, and the underlying fix is verified against the OP's two-panel repro. CI on a clean runner should be the authoritative check.

Notes

• Doc string left as-is; no doc changes needed.
• Release notes not yet added (will follow up once I know the PR number, unless the maintainer would prefer me to add <PR>.fix.md in a follow-up commit here).

AI-assisted by Claude Code.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 79.61%. Comparing base (4ba31e4) to head (3f9fcad).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4137   +/-   ##
=======================================
  Coverage   79.61%   79.61%           
=======================================
  Files         120      120           
  Lines       12786    12789    +3     
=======================================
+ Hits        10179    10182    +3     
  Misses       2607     2607           

Flag	Coverage Δ
hatch-test.low-vers	78.86% <100.00%> (+<0.01%)	⬆️
hatch-test.pre	79.45% <100.00%> (-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/_utils.py	78.99% <100.00%> (+0.13%)	⬆️

... and 1 file with indirect coverage changes