_encode_ugrid mutates BASE_GRID_TOPOLOGY_ATTRS

[rajeeja]

uxarray/io/_ugrid.py:95 binds the module-level dict without copying and then adds keys to it, so every to_xarray("UGRID") / to_netcdf call in a process permanently grows the shared constant. Any grid encoded afterwards advertises variables and dimensions it does not have.

import numpy as np, uxarray as ux
from uxarray.conventions import ugrid

hpgrid = ux.Grid.from_healpix(zoom=1)
hpgrid.edge_lon; hpgrid.face_edge_connectivity
hpgrid.to_xarray("UGRID")
print(len(ugrid.BASE_GRID_TOPOLOGY_ATTRS))  # 11, was 6

minimal = ux.Grid.from_topology(
    node_lon=np.array([0., 1., 1., 0.]),
    node_lat=np.array([0., 0., 1., 1.]),
    face_node_connectivity=np.array([[0, 1, 2, 3]]),
)
ds = minimal.to_xarray("UGRID")
print("n_edge" in ds.dims, ds.grid_topology.attrs["edge_dimension"])  # False n_edge

• Output is non-compliant UGRID: grid_topology points at n_edge, face_lon face_lat, face_edge_connectivity etc. that are not in the dataset.
• Order-dependent, so invisible in single-grid scripts and tests; shows up in notebooks, batch conversions, and long-lived servers.
• The .copy() was added in encode/save to UGRID after healpix creation #1279 and removed in Deprecate Grid.encode_as, Add esmf writer, Add tests, fix to_netcdf bug after healpix mesh creation  #1282 (Deprecate Grid.encode_as, Add esmf writer, Add tests, fix to_netcdf bug after healpix mesh creation  #1282 (comment)) on the reasoning that the added keys were not in the base dict. They are not, until the first call adds them.

Fix is simply grid_topology = dict(ugrid.BASE_GRID_TOPOLOGY_ATTRS) plus a regression test.