Add sharded ocean climate simulation test

[glwagner]

Summary

• Add compile and run scripts for sharded ocean climate simulation using a tripolar grid with the coupled ocean-atmosphere model (sharding/sharded_ocean_climate_simulation_{run,compile}.jl)
• Refactor data_free_ocean_climate_model_init to accept (arch, Nx, Ny, Nz) with grid_type, halo, Δt, and set_initial_conditions kwargs
• Add simple_tripolar_grid helper in model_utils.jl
• Update Compile.yml and Run.yml workflows to include sharded_ocean_climate in the sharded test matrices

This tests sharding with a different grid topology (tripolar vs lat-lon) and atmosphere coupling, complementing the existing sharded_baroclinic_instability test which only uses a simple lat-lon grid.

Known issues

The test currently surfaces three Reactant/Oceananigans limitations with distributed TripolarGrid:

1. set!(field, function) on distributed TripolarGrid fails (unsupported dynamic function invocation in LLVM) — worked around with set_initial_conditions=false
2. ImmersedBoundaryGrid construction fails for distributed TripolarGrid — worked around with grid_type=:simple_tripolar
3. Coupled model compilation hits a Reactant MLIR bug (llvm.call op operand type mismatch in raise_triton_custom_call pass) — needs upstream Reactant fix

Test plan

• CI sharded ocean climate compile job passes
• CI sharded ocean climate run job passes
• Existing sharded baroclinic instability tests still pass
• Existing serial ocean climate tests still pass

🤖 Generated with Claude Code

[glwagner]

cc @romanlee @dkytezab @Pangoraw

[Pangoraw]

OOMs again, should we diminish the problem size like in #258 ?

[glwagner]

OOMs again, should we diminish the problem size like in #258 ?

I think so. Also baroclinic instability -- hm!