Reads multiple zarr files

[Mikejmnez]

Closes #70

There were a couple of permutations with the new added capabilities in xarray. For now, this added code makes it possible for intake to:

1. When a single zarr file needs to be read, it still uses xr.open_zarr, as before, and the get_mapper option for the url_path as before. Another option is to use xr.open_dataset and use fspec.open_local, as it does in the netcdf case. I don't know enough to justify choosing between one or the other (xr.open_zarr is not being deprecated anymore), I just decided to still use xr.open_zarr for simplicity (no code change there).

2. When multiple zarr files, a glob-path like /directoryA/subdir*/subsub*/* can be passed directly to xr.open_mfdataset. In such case, the code makes sure that engine='zarr' is passed as an argument.

In both cases, chunk='auto' is set as the default.

[martindurant]

@Mikejmnez : have you seen pydata/xarray#4461 ?

This solved the same sort of problem here, but at the higher level and with less code. We can definitely include this, at least until the xarray path is concrete.

 >       assert r['datashape'] is None

Datashape is no longer referenced in the intake main package, since it was always None and not used for anything. You can just remove these asserts.

[Mikejmnez]

@martindurant I haven't seen it, thanks for pointing it out to me. Just to understand, with the new (high level) implementations, should url be passed as is into open_dataset? It looks like xarray will then be able to 'handle' the two different options (fs.get_mapper or the glob-path) when appropriate. Is this correct?

[martindurant]

Exactly, xarray should handle all the cases via open_dataset or open_mfdataset. I'm not sure that there is a plan to detect that a URL is glob-like, though.