Fix base mesh plotting with cartopy 0.25 and matplotlib 3.11

[xylar]

Creating gridline labels with draw_labels=True and then hiding some of them (gl.top_labels = False, etc.) confuses cartopy's title placement. GeoAxes._update_title_position() scans gl.top_label_artists whenever gl.geo_labels is true (the default), so the hidden top labels are still included. During a draw, their tight bounding boxes are infinite, so the title ends up at y = inf and the axes tight bounding box becomes non-finite.

The NaN then cascades: tight_layout() gives the axes a NaN position, the "geo" spine path becomes all NaN, and cartopy's gridliner hands NaN vertices to shapely, which raises

GEOSException: IllegalArgumentException: Points of LinearRing do not
form a closed linestring

Instead, only request the labels we actually want. In _plot_cell_width(), that means draw_labels=['left', 'bottom']. In the arrm10to60 and wc14 mesh plots, all four label sets were hidden, so draw_labels=False.

Also drop the redundant fig.canvas.draw() and plt.tight_layout() from _plot_cell_width(), since bbox_inches='tight' already crops the figure. Removing tight_layout() alone is not sufficient: without the label fix, the title is silently dropped and bbox_inches='tight' crops to a garbage bounding box, saving only the colorbar.

List-valued draw_labels requires cartopy 0.22, so add that constraint.

Along the way, satisfy the pre-commit hooks on the arrm10to60 file: isort reorders its imports and flake8 flags a pre-existing F523 (a '{0} x {0}'.format(dlon, dlat) that never used dlat), now an f-string.

This merge also fixes an unrelated provenance issue caused by the transition from conda to pixi.

Checklist

• Document (in a comment titled Testing in this PR) any testing that was used to verify the changes

[xylar]

Testing

I'm running the EC30to60 grid through its paces in:

/lcrc/group/e3sm/ac.xylar/compass_2.0/chrysalis/test_20260725/ec30to60_2

The mesh, WOA23/init and WOA23/performance_test tests passed.