Option to interpret variables using cftime

[TomNicholas]

#68 (comment) explains how we don't currently have support for virtualizing encoded time variables.

To handle this it looks like we need to an another option to open_virtual_dataset, a cftime_vars kwarg, which specifies which variables should be interpreted using cftime. (I think cftime_vars should be a subset of loadable_vars.) This suggestion comes from how kerchunk deals with this problem:

https://fsspec.github.io/kerchunk/reference.html#kerchunk.combine.MultiZarrToZarr

”cf:{var}”, interpret the value of var using cftime, returning a datetime. 
These will be automatically re-encoded with cftime, unless you specify an “M8[*]” 
dtype for the coordinate, in which case a conversion will be attempted.

Originally posted by @dcherian in #73 (comment)

[jsignell]

Just to say that I am taking a look at this.

[jbusecke]

Just wanted to add +1 here, since this is quite important for our work to bring virtual zarr access to ESGF (jbusecke/esgf-virtual-zarr-data-access#8). Please let me know if/where I could help test, and ill try to make some time.

[jsignell]

This is still on my radar and I appreciate the nudge. The test is the one thing that I am confident in 😅 it's more the implementation that needs some thought.

[jsignell]

I'm reading up on encodings now.

[TomNicholas]

It's possible the proper way to implement this is actually to use xarray.decode_cf, which eventually sends variables through these very-relevant-looking CFDateTimeCoder decoding methods

https://github.com/pydata/xarray/blob/56209bd9a3192e4f1e82c21e5ffcf4c3bacaaae3/xarray/conventions.py#L288

see #157 for general idea of using xarray to handle CF conventions.