What do you think about specifying the neighborhood size and weighting separately?

neighbors["size"] = 3 # (or [3, 3]) ?
neighbors["weighting"] = "equal" # (or "sharpen", "gaussian_blur") etc.

I was considering that, but I don't think (or at least I don't know how) all kernels can be set to an arbitrary size. For example, u and v likely only make sense for 3x3. edge and sharpen likely too. gaussian_blur there's likely a formula somewhere, but that's pretty much the only one we can size arbitrarily. I let that for further improvements

I don't know how many can be arbitrary but "equal" (here, "one") certainly can. GRASS has formulas for "gaussian" and "exponential", https://grass.osgeo.org/grass-stable/manuals/r.neighbors.html#:~:text=For%20the%20gaussian%20weighting%20function,*x+y*y))

These wouldn't need to be implemented now, since we can error out for an unsupported combination of kernel and size, but I think the API should support an arbitrary size even if it's only implemented for an equal-weight kernel.

We could also support non-square kernels by padding with zeros to allow e.g. --weighting equal --size 3 5

I've adding a commit adding --size (single value), and combining 'size' and 'kernel' values (with only 'equal' being supported for any size value)