MaskRegions(2d_mask, 3d_grid) fails: xgcm.pad iterates all grid axes

[hdrake]

Symptom

On regionate 0.6.0 (topology-overhaul), passing a 2D boolean mask
(dims=('yh', 'xh')) together with a 3D xgcm.Grid (X, Y, Z axes)
crashes in xgcm.pad:

```
File .../regionate/regions.py:186, in MaskRegions.init
labels, ncomp = connected_components(grid, self.mask)
File .../regionate/boundaries.py:156, in connected_components
Ip = _pad_center(grid, ids_da).transpose(Yc, Xc).values[None]
File .../regionate/boundaries.py:104, in _pad_center
return pad(da, grid, padding_width, padding=padding, fill_value=np.nan)
File .../xgcm/padding.py:869, in pad
da_padded = _pad_basic(data, grid, padding_width, padding, fill_value)
File .../xgcm/padding.py:595, in _pad_basic
_, dim = axis._get_position_name(da)
File .../xgcm/axis.py:240, in Axis._get_position_name
raise KeyError(...)
KeyError: "None of the DataArray's dims ('yh', 'xh') were found in axis coords."
```

_pad_center iterates every axis in grid.axes and calls
xgcm.pad on each; when the Z axis has no matching dim in the 2D id
field, xgcm raises.

Repro (minimal)

Any workflow that traces a 2D isobath / depth mask on a grid that
carries a Z axis — e.g. building a full ice-shelf-aware
xgcm.Grid and then calling
MaskRegions((deptho < 1000.), grid) on the same grid. In our
downstream project (xcryocouple), this exact call pattern worked on
regionate 0.5.5.

Workaround we're using

Building a temporary 2D grid alongside the 3D one, sharing X/Y coords
and padding:

```python
grid2d = xgcm.Grid(
grid._ds,
coords={ax: grid.axes[ax].coords
for ax in grid.axes if ax in ('X', 'Y')},
padding={ax: grid.axes[ax].padding
for ax in grid.axes if ax in ('X', 'Y')},
autoparse_metadata=False,
)
regions = reg.MaskRegions(mask_2d, grid2d)
```

Suggested fix

In connected_components (and any other regionate helper that pads
with _pad_center), only iterate axes that intersect mask.dims.
Or expose an explicit axes=('X','Y') argument on MaskRegions
so a caller can restrict to the axes their mask actually spans.

Happy to open a PR if the fix direction is agreed. Reporting per a
maintainer request to surface friction points during the topology-
overhaul migration.

[hdrake]

Cross-reference: the same bug existed in sectionate.gridutils — padding_width = {ax: (1, 1) for ax in grid.axes} on horizontal-only index arrays, at three sites — and was never found, because the regression test added here (test_dimensionality.py) covers connected_components and MaskRegions but not GriddedRegion, which is the entry point that calls sectionate.grid_section. Filed as MOM6-community/sectionate#52 with a fix in hdrake/sectionate#8, and #30 here adds the missing GriddedRegion case (draft until the sectionate fix lands).

The two tests in this file pass unchanged against stock sectionate 0.4.0rc1, so nothing here was wrong — the coverage just stopped one function short of the library boundary.

_{Investigated with AI assistance (Claude Code).}