Deal with x ⇆ y, λ ⇆ φ, etc when filling halos for metrics/coords in ConformalCubedSphereGrid

[navidcy]

I found that the issue #3254 was because while filling the metrics at the halo regions, we didn't took into consideration that Δx ⇆ Δy in non-trivial cubed sphere connections East ⇆ North.

This PR deals with that.

Closes #3254

[navidcy]

using Oceananigans
using Oceananigans.MultiRegion: getregion
using GLMakie

Nx, Ny = 5, 5
H = 3

range_with_offset(N, H, offset::Int) = (-H + 1 + offset*N):(N + H + offset*N)

grid = ConformalCubedSphereGrid(panel_size = (Nx, Ny, 1), z = (0, 1), horizontal_direction_halo = H)

j_index = 2
i_index = Nx+1 - j_index

fig = Figure(resolution=(2200, 600), fontsize=30)
ax = Axis(fig[1, 1])
lines!(ax, range_with_offset(Nx, H, 0), parent(getregion(grid, 1).Δxᶜᶜᵃ)[:, j_index+H], linewidth=4, color = (:green, 0.9), label="panel 1")
lines!(ax, range_with_offset(Nx, H, 1), parent(getregion(grid, 2).Δxᶜᶜᵃ)[:, j_index+H], linewidth=8, color = (:red, 0.5), label="panel 2")
lines!(ax, range_with_offset(Ny, H, 2), parent(getregion(grid, 4).Δyᶜᶜᵃ)[i_index+H, :], linewidth=4, color = (:blue, 0.5), label="panel 4")
vlines!(ax, [1, Nx+1, 2Nx+1, 3Nx+1], linewidth=8, color=(:black, 0.3))
fig[1, 2] = Legend(fig, ax, framevisible = false)
fig
save("metrictest.png", fig)

while BEFORE THIS PR:

[navidcy]

I wanted to add tests but the current test functionality is difficult to adjust to work for 2D arrays.

Oceananigans.jl/test/test_multi_region_cubed_sphere.jl

Lines 9 to 62 in 39e9b6f

	function get_range_of_indices(operation, index, Nx, Ny)
	if operation == :endpoint && index == :first
	range_x = 1
	range_y = 1
	elseif operation == :endpoint && index == :last
	range_x = Nx
	range_y = Ny
	elseif operation == :subset && index == :first # here index is the index to skip
	range_x = 2:Nx
	range_y = 2:Ny
	elseif operation == :subset && index == :last # here index is the index to skip
	range_x = 1:Nx-1
	range_y = 1:Ny-1
	else
	range_x = 1:Nx
	range_y = 1:Ny
	end
	return range_x, range_y
	end
	function get_halo_data(field, side, k_index=1; operation=nothing, index=:all)
	Nx, Ny, _ = size(field)
	Hx, Hy, _ = halo_size(field.grid)
	range_x, range_y = get_range_of_indices(operation, index, Nx, Ny)
	if side == :west
	return field.data[-Hx+1:0, range_y, k_index]
	elseif side == :east
	return field.data[Nx+1:Nx+Hx, range_y, k_index]
	elseif side == :south
	return field.data[range_x, -Hy+1:0, k_index]
	elseif side == :north
	return field.data[range_x, Ny+1:Ny+Hy, k_index]
	end
	end
	function get_boundary_indices(Nx, Ny, Hx, Hy, side; operation=nothing, index=:all)
	range_x, range_y = get_range_of_indices(operation, index, Nx, Ny)
	if side == :west
	return 1:Hx, range_y
	elseif side == :south
	return range_x, 1:Hy
	elseif side == :east
	return Nx-Hx+1:Nx, range_y
	elseif side == :north
	return range_x, Ny-Hy+1:Ny
	end
	end

I'd like to rewrite those to use the connectivity property of the ConformalCubedSphereGrid rather than a series of if statements. Also address ##3242. But I wasn't planning to do this on this PR.

[glwagner]

So the problem is across panel 2 and 4? And there is also a problem across panel 1 and 2, but oddly the coordinates were still continuous originally.