Only call set_precomputed_quantities! once in the implict stage

[simonbyrne]

Currently we call set_precomputed_quantities! in the Wfact!:

ClimaAtmos.jl/src/prognostic_equations/implicit/wfact.jl

Line 64 in 17c5a2f

set_precomputed_quantities!(Y, p, t)

and once again in the implicit_tendency!:

ClimaAtmos.jl/src/prognostic_equations/implicit/implicit_tendency.jl

Line 12 in 17c5a2f

set_precomputed_quantities!(Y, p, t)

But the state should be the same in both cases, so this should be unnecessary.

[simonbyrne]

This appears to break the SSP tests, since they don't quite follow the same logic.

I think the general solution here is to move set_precomputed_quantities! so that it is applied after each stage update: in this way, our cache will always be in sync with the current state (which means that we don't need to call it in the diagnostics). This will mean that we don't have to do it in diagnostics.

[simonbyrne]

We also call it twice in the explicit stage: once in limited_tendencies!:
https://github.com/CliMA/ClimaAtmos.jl/blob/f751e691c05ae935ff30d82b7ef804021f5355ea/src/prognostic_equations/limited_tendencies.jl

and once in remaining_tendencies!:

ClimaAtmos.jl/src/prognostic_equations/remaining_tendency.jl

Line 7 in f751e69

set_precomputed_quantities!(Y, p, t)

With the ARS343, the second is called directly after the first. Can we safely get rid of it in the second one?

[simonbyrne]

@dennisYatunin ?

[dennisYatunin]

Yes, the two explicit tendencies are computed for the same state, so their precomputed quantities should be identical. The order in which the tendencies are evaluated is implementation-dependent, though, so it would definitely be safer to separate out the precomputed quantities computation and guarantee that it will only be called once.