Remove tendency computation at model construction

[simone-silvestri]

At the moment, the tendencies were updated at each call of update_state!. This PR makes it so that the tendencies are computed only inside the time-stepping where we need them.

see ClimaOcean#164

[glwagner]

The logic should be reversed, we should have to include "compute_tendencies=false" as an optimization. Rather than what's implemented here, which does the unsafe thing by default. The problem with this logic is that it makes it harder to implement new models. The optimizations should be the optional thing basically, if one is naive, then things should work even if they are slower than they could be

[simone-silvestri]

I just noticed that the SingleColumnGrid was never computing the tendencies. Is this a bug? Was this mode working? I can add a test for it here

[simone-silvestri]

Found out the issue, because of this line

Oceananigans.jl/src/Models/HydrostaticFreeSurfaceModels/update_hydrostatic_free_surface_model_state.jl

Lines 28 to 30 in 5a3d3c9

	update_state!(model::HydrostaticFreeSurfaceModel, callbacks=[]; compute_tendencies = true) =
	update_state!(model, model.grid, callbacks; compute_tendencies)

the update_state! in this file was neglected because of the lack of the compute_tendencies keyword argument. So, apparently, it is a bug, but one that only influences performance and not results

[simone-silvestri]

This should be ready to merge