Flux partitioning

[LenkaNovak]

Leaf issue of and closes #311

Currently, fluxes are calculated on the atmospheric grid. If an atmospheric grid cell overlays two surfaces (e.g. land and ocean), the states of those surfaces are averaged into "combined fields", which are computed once in each cell. We want to add the ability to calculate fluxes in the coupler more accurately over different surfaces separately.

Components

• add such functionality into FluxCalculator
  • partitioned_turbulent_fluxes! iterates over all surface models and calculates their individual fluxes. The area-weighted average is then passed to the atmosphere
• add unit tests
• implement in the AMIP driver
• add bucket-specific surface thermo state

QA

• Ensure energy conservation (n.b. some minor physical behavior expected)

• Ensure all buildkites run (or note those which are unstable with this more refined flux calculation)
  • coarse_single_modular fails after 7 days (new stability region to be found as part of O1.6.2.2 (coupler) Physical Components for ClimaCoupler v0.1.0 Release #359 )
  • perfomance tests fail due to 60x more allocations, translating into 30% increase in walltime. Reduced performance is not unexpected due to the fact that this methods triples the flux calculation by design, but also due to the current Atmos interface hacks. Following improvements in Amtos-Coupler interfacing, as outlined in O5.1.4 (coupler) Software Revamp for ClimaCoupler v0.1.0 Release #358 , the allocations should be reduced to a more reasonable value.