Support Oceananigans v0.95 and bump v0.13.4

[ali-ramadhan]

🤞 that everything works out of the box!

[ali-ramadhan]

I think I'm seeing a couple of different errors pointing to this function

function intercept_tracer_tendencies!(model, intercepted_tendencies)
    for (name, field) in enumerate(intercepted_tendencies)
        field .= Array(interior(model.timestepper.Gⁿ[name + 3]))
    end
end

in test_sediments.jl. Something about name + 3 being out of bounds, and something about the array dimensions not matching depending on the test.

I'm skipping the sediment tests for now just to see if the rest of the tests pass on CI (they did on my machine).

@jagoosw I hope you don't mind that I re-organized test_sediments.jl a bit while investigating.

[jagoosw]

That is quite strange, the changes since 0.93 shouldn't have effected any of this right?

[ali-ramadhan]

Yeah I agree and I couldn't find any hints in the diffs or the changelog... I'll try to take a closer look later today.

[ali-ramadhan]

Because of CliMA/Oceananigans.jl#3930 I had to rename ab2_step_free_surface! to step_free_surface! .

I think there's a wider discussion to be had around supporting hydrostatic RK3 time stepping with OceanBioME.jl but I'll open a new issue for it.

[ali-ramadhan]

Hmmm, new error seems to be related to store_flat_tendencies! going out of bounds for sediment models.

[ali-ramadhan]

Let's try this again.

Right now we're depending on the main branch of Oceananigans.jl until a new version is tagged (hopefully v0.95.0 soon).

Docs are failing because they're pulling in v0.92.x based on the [compat] entries and there's no way to enforce a specific branch via [compat] so I added a Manifest.toml to docs/ to get the tests to pass. The aim is to remove this Manifest.toml once we can upgrade to v0.95.0.

[ali-ramadhan]

I think we're also encountering CliMA/Oceananigans.jl#3964 here.

[ali-ramadhan]

Updating to Oceananigans v0.95 now!