Should we test also models run with tuple of closures and/or using different models?

[navidcy]

E.g., in addition of this:

Oceananigans.jl/test/test_time_stepping.jl

Lines 25 to 40 in 057e25c

	function time_stepping_works_with_closure(arch, FT, Closure; buoyancy=Buoyancy(model=SeawaterBuoyancy(FT)))
	# Add TKE tracer "e" to tracers when using CATKEVerticalDiffusivity
	tracers = [:T, :S]
	Closure === CATKEVerticalDiffusivity && push!(tracers, :e)
	# Use halos of size 2 to accomadate time stepping with AnisotropicBiharmonicDiffusivity.
	grid = RegularRectilinearGrid(FT; size=(1, 1, 1), halo=(2, 2, 2), extent=(1, 2, 3))
	model = NonhydrostaticModel(grid=grid, architecture=arch,
	closure=Closure(FT), tracers=tracers, buoyancy=buoyancy)
	time_step!(model, 1, euler=true)
	return true # Test that no errors/crashes happen when time stepping.
	end

we also have tests using, e.g., closure = (AnisotropicDiffusivity(FT), Closure(FT)) and also some tests using, e.g., HydrostaticFreeSurfaceModel?

[navidcy]

(The backstory of this is that I wasn't able to run HydrostaticFreeSurfaceModel with two closures on the GPU...)

[glwagner]

Pretty sure we do test closure tuples