Unreliable remote I/O w/ open_restart

[nbren12]

Lately, we had hanging tests on fv3net due to I/O problems with fv3gfs.util.open_restart. See ai2cm/fv3net#951. This only happened in the context of an FV3 run. open_restart worked fine with a simple testing script.

This PR solved these problems by using fsspec.get to pre-download the restart files: ai2cm/fv3net#978.

My guess is that it's related to multiple MPI jobs launching downloads and getting stuck. I've also found that fsppec.open does many more HTTP requests than fsspec.get. Maybe we should switch to use that here.

Since remote I/O paradigms are in constant flux, it might be simpler to deprecate this remote I/O capability, and focus fv3gfs-util on local usage.