numerical values for time (instead of datetime objects) are erroneously converted

[aportagain]

Numerical values for time (instead of datetime objects) are erroneously converted. We should probably either only accept datetime objects, or for numerical values preserve the units. The example currently in README.md even shows this in action:

In [1]: import json
   ...: import pprint as pp
   ...: import xarray as xr
   ...: from cfjson.xrdataset import CFJSONinterface
   ...: dset = xr.Dataset(data_vars={'hmo': ('time', [0.5, 0.6, 0.7])}, coords={'time': [1, 2, 3]})
   ...: dset['hmo'].attrs = {'standard_name': 'sea_surface_wave_significant_height', 'units': 'm'}
   ...: dset['time'].attrs = {'standard_name': 'time', 'units': 'days since 2020-01-01'}

In [2]: pp.pprint(json.loads(dset.cfjson.json_dumps()))
{'attributes': {},
 'dimensions': {'time': 3},
 'variables': {'hmo': {'attributes': {'standard_name': 'sea_surface_wave_significant_height',
                                      'units': 'm'},
                       'data': [0.5, 0.6, 0.7],
                       'shape': ['time']},
               'time': {'attributes': {'units': 'ISO8601 timestamps'},
                        'data': ['1970-01-01T00:00:00Z',
                                 '1970-01-01T00:00:00Z',
                                 '1970-01-01T00:00:00Z'],
                        'shape': ['time']}}}

In [3]: