GaussianMask doesn't work with Flat directions

[glwagner]

MWE:

using Oceananigans

grid = RectilinearGrid(size = (10, 10),
                       x = (0, 1),
                       z = (0, 1),
                       topology = (Periodic, Flat, Bounded))

mask = GaussianMask{:x}(center=1, width=0.1)
sponge = Relaxation(rate=1; mask)

model = HydrostaticFreeSurfaceModel(; grid, forcing = (; u=sponge))

simulation = Simulation(model; Δt=1, stop_iteration=1)
run!(simulation)

The fix is a little involved because we can't distinguish between the arguments (y, z) and (x, y) by function arguments alone. Thus the mask itself needs to have grid information.

This and other bug are the extended consequences of changing node to drop flat dimensions (#3355, with discussion on #3285). I still think the price was worth it for betting scripting, but it has come with a pretty significant trade-off in terms of under the hood code complexity...