Speed up landice gridded_flood_fill with scipy.ndimage

[trhille]

Speed up gridded_flood_fill with scipy.ndimage instead of nested loops. For the antarctica_8km_2024_01_29.nc, this reduced the time from about 7s to 0.005s. This could make it much more manageable to use high-resolution gridded data sets to determine cell spacing for large meshes.

Checklist

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

[trhille]

Testing

When using the antarctica_1km_2024_01_29.nc source file, the current implementation of landice.mesh.gridded_flood_fill in main takes 495.47 seconds. With these changes, it takes 0.32 seconds.
@xylar @matthewhoffman, I think you'll like this :)

I still need to confirm that these result in similar masks.

[trhille]

Here are results for the 3–30km GIS mesh, using the head of main (left) and the head of this branch (right). Results Both meshes have nCells = 54173 and areaCell fields are identical between the two. Using this branch, gridded_flood_fill took 0.0086 seconds; using main it took 7.55 s.

[matthewhoffman]

@trhille , this looks great, and thanks for adding the test showing no answers change. CI is failing for some unrelated reason, so once I sort that out, I will merge.

[matthewhoffman]

@xylar , it looks like the CI is failing when trying to build jigsaw. I forced the CI to rerun and got the same error. I'm confused because this branch is based on main, and the CI appeared to pass on the most recent PR. I can dig deeper, but wanted to check if you had any insight into this, as I know it happens periodically.

[xylar]

The way of building jigsaw changes (or can change) in #944. I don't really want to debug the "old" way of building jigsaw.

[matthewhoffman]

Based on group discussion, merging without CI passing.