[FEA] Support grouping by a Series in dask_cudf groupby

[shwina]

Grouping by an external Series doesn't always work in dask_cudf:

In [34]: df = dask_cudf.from_cudf(cudf.DataFrame({'a': [1, 2, 3, 4, 5]}), npartitions=1)
In [35]: s = dask_cudf.from_cudf(cudf.Series([1, 1, 1, 2, 2], name='id'), npartitions=1)
In [36]: df.groupby([s]).agg(["sum"]).compute() # error

...

ValueError: Metadata inference failed in `eq`.

Original error is below:
------------------------
TypeError("cannot broadcast <class 'str'>")

Although, for very simple aggregations, it does -- note how I'm not wrapping "sum" in a list:

In [37]: df.groupby([s]).agg("sum").compute()
Out[37]:
    a
id
2   9
1   6