Cdflib

Fixing a bug that occurs in cdf_to_xarray when DEPEND_0 has a length of 1

Allowing lists of numpy datetime64 objects to appear in variable attributes

Making it more explicit that attributes given a "None" value will be skipped.

Minor change to xarray_to_cdf:
* Convert FILLVAL to the same data type as the primary data when "istp=True" is specified

cdfwrite

* Added some support for casting numbers correctly if they have different inputs to the CDF. For example, if CDF_REAL8 is specified as the CDF type to use but the data is given as an integer, the data will be cast to a float64 before being written to the CDF.

* In the xarray_to_cdf function, added the datetime64_to_cdftt2000 to flag to convert np.datetime64 to cdftt2000
* in the xarray_to_cdf function, made a change to the datetime converter than was causing the data to be cast as a float, losing a little precision when using datetime64 ints
* Added unit tests for the above improvements, as well as the issue from 1.1.2 with arrays of numpy strings being read in as variable data