NCDataset.from_json() breaks if a variable doesn't have the "type" attribute

[aportagain]

NCDataset.from_json() breaks if a variable doesn't have the "type" attribute, but our current spec doesn't say that "type" is required in general, only to ensure reversible transformations. I talked to Dave about this and we decided that ideally we should make it work according to the current spec. But first for my specific use case, I'll probably open a separate issue for missing "type" support in the xrdataset.CFJSONinterface.json_dumps().

To reproduce:

from cfjson.ncdataset import NCDataset
nc = NCDataset('/tmp/foo.nc', 'w')
j = '\n{\n  "dimensions": {\n    "time": 1\n  },\n  "attributes": {},\n  "variables": {\n    "time": {\n      "attributes": {\n        "units": "ISO8601 timestamps"\n      },\n      "shape": [\n        "time"\n      ],\n      "data": [\n        "2017-01-01T00:00:00Z"\n      ]\n    },\n    "hs": {\n      "attributes": {},\n      "shape": [\n        "time"\n      ],\n      "data": [\n        3.058\n      \n      ]\n    }\n  }\n}\n'
nc.from_json(j)

returns this error:

Issue occured when inserting global attributes
Could not create variable time
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/source/cf-json/cfjson/ncdataset.py", line 147, in from_json
    self.createVariable(var, dico['variables'][var]['type'],
KeyError: 'type'