focal_stats and hotspots: result .name differs across backends

[brendancol]

Summary

The .name of the DataArray returned by focal_stats and hotspots is not the same across the four backends. On the dask paths it leaks an internal dask graph token.

Details

focal_stats:

• numpy / dask+numpy: 'focal_apply' (inherited from the internal apply call)
• cupy: None
• dask+cupy: a dask graph token, e.g. '_trim-d142fa3d...'

hotspots:

• numpy / cupy: None
• dask+numpy / dask+cupy: a dask graph token, e.g. '_trim-b75965c2...'

Root cause: the dask paths build the output xr.DataArray without an explicit name=. When the backing array is a dask array, xarray adopts the dask array's internal graph-layer name as the public .name. That token changes from call to call and exposes an implementation detail. The numpy/cupy paths leave the name as None (or inherit a different value).

.name is user-visible: it shows in the repr, becomes the variable key in to_dataset(), and sets default plot labels. So a chained pipeline ends up with a different name depending on which backend ran.

Fix

Set a stable .name on the returned DataArray in both functions so all four backends agree. focal_stats keeps the 'focal_apply' name the numpy path already emits; hotspots uses 'hotspots'.

The rest of the metadata (attrs res/crs, coords, dims) is already preserved consistently across all four backends. Only .name was off.