Implicit stencil test time regression

[simonbyrne]

Describe the bug

#767 appears to have caused the implicit stencil tests to go from ~20 mins to ~60 mins. This is probably caused by the DataLayouts changes

[charleskawczynski]

I've got a wip file where I was experimenting/digging around to find out what's wrong.

[charleskawczynski]

Along with:

julia -O0 --check-bounds=yes --project=test
julia --project=test

push!(ARGS, "--float_type", "Float32")
include(joinpath("test", "Operators", "finitedifference", "implicit_stencils_utils.jl"))

(s, parsed_args) = parse_commandline();
const FT = parsed_args["float_type"] == "Float64" ? Float64 : Float32

function do_work(all_ops)
    for _ in 1:100
        test_pointwise_stencils_apply(all_ops)
        # test_pointwise_stencils_compose(all_ops)
    end
end

center_space = get_space(FT);
all_ops = get_all_ops(center_space);

do_work(all_ops) # compile first, toggle to see compile time allocations
import Profile
import ClimaCore
Profile.clear_malloc_data()
Profile.clear()
prof = Profile.@profile do_work(all_ops)
import ProfileCanvas
ProfileCanvas.html_file(joinpath(pkgdir(ClimaCore), "implicit_stencil_flame.html"))

Requires #1229.

[charleskawczynski]

#1305 seems to have fixed this!