Distributed LocStream problem at C2880 resolution on large processor-count

[metdyn]

I remember watching the screen together with @atrayano for an AIST run testing MAPL Trajectory sampler with 21600 cores (180 nodes) for C2880 runs. The code froze for 30 minutes to generate a distributed LocStream using a cubed sphere background with C2880.

Later on @bena-nasa illustrated the problem with a test code using a few hundred line calling MAPL and ESMF: the code slow down is related to
locstream = ESMF_LocStreamCreate(locstream,background=grid,rc=status)

https://github.com/GEOS-ESM/MAPL/blob/feature/ygyu/locstream_chunk/base/MAPL_LocStreamFactoryMod.F90#L92-L95