use TripolarGrid for Oceananigans sim

[juliasloan25]

Purpose

Use a TripolarGrid instead of a capped lat/lon grid when constructing an Oceananigans simulation.

closes #1382

Note: Oceananigans.Fields.interpolate doesn't work for the TripolarGrid. We use this to remap Oceananigans fields from the OC grid to the ClimaCore field, so we can't switch to TripolarGrid until Oceananigans has a method that can regrid the TripolarGrid. This should be added in CliMA/ClimaOcean.jl#463 - requires Oceananigans v0.99.2.

Note 2: Oceananigans is now using XESMF.jl to regrid for their TripolarGrid. This PR uses XESMF bilinear regridding to go from the Oceananigans grid to the ClimaCore boundary space. We continue to use ClimaCore.Remapping to go from ClimaCore to Oceananigans because it uses more information about the spectral element space than XESMF is able to. See this link for information about the limitations of using XESMF with spectral element spaces.

To-do

• construct TripolarGrid instead of LatitudeLongitudeGrid
• store remapping objects for both directions in the OceananigansSimulation object (Oceananigans -> ClimaCore and ClimaCore -> Oceananigans)
• add a function Interfacer.get_remapper_to_cc that retrieves the remapping object to get from a component model's grid to ClimaCore (this is nothing for components that use ClimaCore, but for other components it's necessary to have access to the remapper object when remap is called)

Additional notes

• TripolarGrid constructs a OrthogonalSphericalShellGrid, which also uses lat/lon, so we can still use the same ClimaCore remapper to go from ClimaCore to Oceananigans
• exchanged fields are T, S, LHF, SHF, u, v - only u and v are vectors, and these are already rotated onto the grid in set_from_extrinsic_vectors!

[glwagner]

@juliasloan25 how do I run the coupled simulation ? Is there a way to run it outside the caltech cluster (eg on a GPU system we have at MIT)?

[juliasloan25]

@juliasloan25 how do I run the coupled simulation ? Is there a way to run it outside the caltech cluster (eg on a GPU system we have at MIT)?

Sorry I just saw this - I get a lot of Github notifications so they get missed often. We have a docs page on running the coupled simulation. It shows how to run interactively or from the command line. You should be able to run it locally or on a different cluster, and it'll pick up the device to run on automatically (i.e. if a GPU is available it will run on GPU). Let me know if there are problems.

[juliasloan25]

Suggested change

	OceananigansXESMFExt.xesmf_coordinates(grid, OC.Center(), OC.Center(), OC.Center())
	XESMF.xesmf_coordinates(grid, OC.Center(), OC.Center(), OC.Center())