Better zarr store update from set of nc files

[remicousin]

          In https://bitbucket.org/iridl/subx-cloud/src/master/zarrify.py I took the approach of creating an empty store and then writing into regions of it. (The motivation in that case was to parallelize writes.) It might be more complicated that what's needed in this case.

It would be nice to be able to either (a) skip opening any netcdf files whose data are already present in the zarr, for speed, or (b) compare pre-existing zarr data to what's in the netcdf files, for extra safety when you're not in a hurry. But that can wait for a subsequent version.

Originally posted by @aaron-kaplan in #404 (review)