[BUG] (De-)Serialization Review

[quasiben]

In submitting a PR fixing index serialization: #2610, @kkraus14 found additional

@quasiben Another potential serialization / deserialization issue: https://github.com/rapidsai/cudf/blob/branch-0.10/python/cudf/cudf/groupby/groupby.py#L69
There's a Series which is backed by device memory underneath the groupby object.

@quasiben Another potential serialization / deserialization issue: https://github.com/rapidsai/cudf/blob/branch-0.10/python/cudf/cudf/groupby/groupby.py#L69
There's a Series which is backed by device memory underneath the groupby object.
This one too: https://github.com/rapidsai/cudf/blob/branch-0.10/python/cudf/cudf/groupby/groupby.py#L140

We should ensure the correctness of the serialization scheme by adding additional tests to test_serialize.py. This could be done by adding more "flavors" of cudf objects, confirming data is GPU backed after serialization, confirming data is in the frames of the serialized return and not the header

[quasiben]

I'm going to close this. @shwina and I have been through the code a number of time now and #2619 address this issue as well