Add compression filters

[ph-kev]

Examining the checkpoints that are saved by ClimaCoupler-AMIP buildkite run, the biggest cache files are those saved by JLD2.

JLD2 supports compression which can significantly reduce file size. A shuffle filter is applied, because it exploits the patterns in the higher-order bytes, given that the data is from a climate simulation. See the example in HDF5.jl (note that both HDF5 and JLD2 shuffle filters work the same way). Then, a ZstdFilter filter is applied with level 1 to prioritize speed over compression.

➜  checkpoints du -sh *
1.3M    checkpoint_BucketSimulation_94867200.hdf5
67M     checkpoint_cache_1_BucketSimulation_94867200.jld2
54G     checkpoint_cache_1_ClimaAtmosSimulation_94867200.jld2
19M     checkpoint_cache_1_PrescribedIceSimulation_94867200.jld2
21M     checkpoint_cache_1_PrescribedOceanSimulation_94867200.jld2
42M     checkpoint_ClimaAtmosSimulation_94867200.hdf5
5.3M    checkpoint_coupler_fields_1_94867200.jld2
144K    checkpoint_PrescribedIceSimulation_94867200.hdf5

TODO

• Make a small benchmark to compare performance between no compression and different compression filters (time to checkpoint and size of checkpoints)