Preallocate thermal state

[Sbozzolo]

Fixes #588

Leads to x2 improvement:

https://buildkite.com/clima/climaland-benchmark/builds/65#018f3174-1371-42ee-8dc2-eca52228dc0c

Average time: 5.3 s

https://buildkite.com/clima/climaland-benchmark/builds/53#018f3104-e10b-4e7e-af80-0a2476646201

Average time: 10 s

[kmdeck]

I think we should discuss where the ts is allocated/stored. In this particular case, I think it could make sense to store it/add it to p in:

• p.drivers.ts_atmos, and update not with evaluate! like the other driver variables, but instead update with the set_atmos_ts! function you wrote. We already have an update_drivers function which this would be easy to add to. This is consistent also in that it will update only update when update_drivers is called (as a callback), which is when the other drivers are updated, and not every timestep/stage (even though the drivers may not be changing each step).

More generally - if we need to add scratch variables - we can do so pretty easily via our existing auxiliary_vars and boundary_vars functionality. We may already have some, actually. then they would be stored in p.soil.scratch (e.g.) and we can choose which models need them/for which boundary conditions. We could also make a new function like you did and add scratch variables for a model to p.scratch, like we do for driver variables. But that is a little bigger of a job and I dont think it is needed for this PR.

[kmdeck]

if certain boundary conditions or source terms require scratch, but others dont, we could make a scratch_vars(model) = (scratch_vars(model.boundary_conditions)..., scratch_vars(model.sources)...) function.

[juliasloan25]

This looks great! I agree with Kat's comment, and it looks like those changes have already been incorporated.

Can we fully remove the construct_atmos_ts function and remaining calls to it now? (e.g. in exp/Standalone/Soil/evaporation.jl, and a couple of the tests)

[kmdeck]

This looks good to me, thank you! nice improvement to the code.