[FEA] Support "COLLECT" aggregation on struct columns in cuDF

[GregoryKimball]

Is your feature request related to a problem? Please describe.
I would like to perform COLLECT aggregations on Struct columns. Based on initial triage by @shwina , adding COLLECT to _STRUCT_AGGS in groupby.pyx would be the first step.

Describe the solution you'd like
I would like groupby.agg(list) to be supported for Struct columns.

Describe alternatives you've considered
I couldn't find a way to construct a nested List<Struct> without doing a device-host roundtrip and assembling a new cudf.DataFrame. If there are other tools we could use to pack a Struct column into a List column I would love to know.

Additional context
Here is an example of packing a Struct column into a List column.

>>> df = cudf.DataFrame(index=[0,0,1], data={'a':[
...    {'k':'v1'}, {'k':'v2'}, {'k':'v3'}
... ]})
>>> df
             a
0  {'k': 'v1'}
0  {'k': 'v2'}
1  {'k': 'v3'}
>>> df.groupby(level=0).agg(list)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/conda/envs/rapids/lib/python3.8/contextlib.py", line 75, in inner
    return func(*args, **kwds)
  File "/opt/conda/envs/rapids/lib/python3.8/site-packages/cudf/core/groupby/groupby.py", line 458, in agg
    ) = self._groupby.aggregate(columns, normalized_aggs)
  File "groupby.pyx", line 295, in cudf._lib.groupby.GroupBy.aggregate
  File "groupby.pyx", line 185, in cudf._lib.groupby.GroupBy.aggregate_internal
pandas.core.base.DataError: All requested aggregations are unsupported.
>>> expected = cudf.DataFrame({'a':[
...    [{'k':'v1'}, {'k':'v2'}],[{'k':'v3'}]
... ]})
>>> expected
                            a
0  [{'k': 'v1'}, {'k': 'v2'}]
1               [{'k': 'v3'}]

[GregoryKimball]

Note, simply adding the COLLECT aggregation works to create a List column, but the full dtype is not preserved.

>>> df = cudf.DataFrame(index=[0,0,1], data={'a':[
...    {'k':'v1'}, {'k':'v2'}, {'k':'v3'}
... ]})
>>> df.groupby(level=0).agg(list)
                            a
0  [{'0': 'v1'}, {'0': 'v2'}]
1               [{'0': 'v3'}]

See 11687, 11671.

[ttnghia]

Sorry is this FEA for Python or Java?
We have the collect_list aggregation and I'm still not clear on what is difference from COLLECT here?

[wence-]

Pretty sure this is for Python (have updated labels); as @GregoryKimball notes if you add "COLLECT" to the set of supported aggregations here

cudf/python/cudf/cudf/_lib/groupby.pyx

Line 61 in 387192c

_STRUCT_AGGS = {"CORRELATION", "COVARIANCE"}

then libcudf will generate columns with the correct data. However, since struct dtype field names are managed in Python the field names are lost. Similar to #11687, I think this can be fixed by transferring relevant dtype information over post-hoc.