Improve coverage of dask-cudf's groupby aggregation, add tests for dropna support

[charlesbluca]

This PR does the following:

• Make sure that all of dask-cudf's SUPPORTED_AGGS have an overriding method for upstream Dask's series / dataframe groupby methods
• Add tests comparing dask-cudf's dropna support to upstream Dask's, as at the moment we are only comparing against cuDF
• Fix the resulting failures of these changes (by properly parsing self.dropna in dask-cudf's groupby code)

As a side note, I think that a larger rethinking of dask-cudf's groupby would pay off well, as currently it seems like we have some "duplicate" tests and aren't really able to discern if groupby_agg was called for a supported aggregation

[charlesbluca]

rerun tests

[charlesbluca]

rerun tests

[codecov]

Codecov Report

Merging #10449 (12c558b) into branch-22.06 (4913a9b) will decrease coverage by 0.12%.
The diff coverage is 100.00%.

@@               Coverage Diff                @@
##           branch-22.06   #10449      +/-   ##
================================================
- Coverage         86.42%   86.30%   -0.13%     
================================================
  Files               143      143              
  Lines             22493    22631     +138     
================================================
+ Hits              19440    19532      +92     
- Misses             3053     3099      +46     

Impacted Files	Coverage Δ
python/dask_cudf/dask_cudf/groupby.py	97.36% <100.00%> (+0.69%)	⬆️
python/dask_cudf/dask_cudf/tests/test_groupby.py	100.00% <100.00%> (ø)
...thon/dask_cudf/dask_cudf/tests/test_distributed.py	18.86% <0.00%> (-67.93%)	⬇️
python/cudf/cudf/io/parquet.py	90.47% <0.00%> (-2.23%)	⬇️
python/dask_cudf/dask_cudf/backends.py	85.51% <0.00%> (-0.94%)	⬇️
python/cudf/cudf/core/column/categorical.py	89.37% <0.00%> (-0.61%)	⬇️
python/cudf/cudf/core/column/decimal.py	90.60% <0.00%> (-0.50%)	⬇️
python/cudf/cudf/core/column/lists.py	91.66% <0.00%> (-0.42%)	⬇️
python/cudf/cudf/core/column/string.py	88.78% <0.00%> (-0.31%)	⬇️
python/cudf/cudf/core/index.py	92.06% <0.00%> (-0.25%)	⬇️
... and 9 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 4913a9b...12c558b. Read the comment docs.

[rjzamora]

Thanks for working on this @charlesbluca! Most of my comments are just minor style suggestions.

[rjzamora]

What is the reason to include those "by" parameters if we will always be skipping them? Where these tests only meant to be skipped when dropna=True?

[charlesbluca]

Good catch - IIRC they don't work when dropna=False, but I will try things out locally to double check

[charlesbluca]

Running the tests locally, it looks like there's a few separate (but likely related) issues in Dask CPU going on:

• multi-column groupbys grouping on categorical columns don't work at all
• dropna=False doesn't work on any multi-column groupbys
• dropna=False doesn't work on single-column groupbys on categorical columns

I think maybe instead of skipping here, we should xfail the currently failing cases - I suggest this so we can track the ongoing work in Dask CPU and remove the xfail once these tests are passing, without having to substantially change the test parameter. How do you feel about this? If you would rather just remove the categorical column cases for now and add a TODO to add those in later on when the dask issue is resolved, I am okay with that 🙂

[rjzamora]

Sorry for mising this 26 days ago @charlesbluca ! I do think xfail makes a lot more sense than skipping.

[charlesbluca]

@gpucibot merge