Do not restrict bathymetry

[simone-silvestri]

With CliMA/Oceananigans.jl#4074 we can do domains that cross the 0 in longitude, so we do not really need to restrict the bathymetry.

TODO: after merging the Oceananigans PR, show that we can do a Mediterranean simulation that includes Gibraltar strait.

[glwagner]

I put the restricting in as a performance consideration, not because we are using different coordinate systems.

[simone-silvestri]

Right, but this restriction limits us from doing stuff like:

julia> grid = LatitudeLongitudeGrid(size = (1000, 500, 100),
                                    latitude  = (0, 60),
                                    longitude = (-15, 50),
                                    z = (-6000, 0))
1000×500×100 LatitudeLongitudeGrid{Float64, Bounded, Bounded, Bounded} on CPU with 3×3×3 halo and with precomputed metrics
├── longitude: Bounded  λ ∈ [-15.0, 50.0]  regularly spaced with Δλ=0.065
├── latitude:  Bounded  φ ∈ [0.0, 60.0]    regularly spaced with Δφ=0.12
└── z:         Bounded  z ∈ [-6000.0, 0.0] regularly spaced with Δz=60.0

julia> bottom = regrid_bathymetry(grid, major_basins=Inf)
1000×500×1 Field{Center, Center, Nothing} reduced over dims = (3,) on LatitudeLongitudeGrid on CPU
├── grid: 1000×500×100 LatitudeLongitudeGrid{Float64, Bounded, Bounded, Bounded} on CPU with 3×3×3 halo and with precomputed metrics
├── boundary conditions: FieldBoundaryConditions
│   └── west: ZeroFlux, east: ZeroFlux, south: ZeroFlux, north: ZeroFlux, bottom: Nothing, top: Nothing, immersed: ZeroFlux
└── data: 1006×506×1 OffsetArray(::Array{Float64, 3}, -2:1003, -2:503, 1:1) with eltype Float64 with indices -2:1003×-2:503×1:1
    └── max=4161.45, min=-5636.4, mean=-101.856

julia> heatmap(bottom)

which is possible on this branch thanks to the new interpolation routine.
It might be possible to try to circshift to limit the bathymetry to the region of interest if we work on it a bit, but I think it's not warranted given that the cost of regridding the bathymetry scales with the grid we interpolate on, not really the grid we interpolate from and the whole native bathymetric data is loaded anyways.