Chunking strategy

[mgrover1]

I figured it would be helpful to track the chunking strategy for these different variables/datasets.

Following NCAR/cesm-lens-aws#34, it looks like chunking by time, with all vertical dimensions and lat/lon in the same chunk would be best for the 3D vars. Here's an example of a 3D monthly pop output:

'chunks': {'time': 6,
                  'z_t': -1,
                  'nlon': -1,
                  'nlat': -1}

Based on the first few tests, it seems like we can use the same chunking strategy as LENS1

[andersy005]

It might be worth trying out a different, moderate chunking scheme (e.g. chunking in nlat, nlon, and z_t as well). I hope that this moderate chunking scheme allows us to meet most use cases in the middle without sacrificing the performance