add min_periods, ddof to groupby covariance, & correlation aggregation

[karthikeyann]

Addresses part of #8691
Add min_periods and ddof parameters to libcudf groupby covariance and Pearson correlation (python needs this)

[codecov]

Codecov Report

Merging #9492 (d6ebfb3) into branch-21.12 (ab4bfaa) will decrease coverage by 0.11%.
The diff coverage is n/a.

@@               Coverage Diff                @@
##           branch-21.12    #9492      +/-   ##
================================================
- Coverage         10.79%   10.67%   -0.12%     
================================================
  Files               116      117       +1     
  Lines             18869    19714     +845     
================================================
+ Hits               2036     2104      +68     
- Misses            16833    17610     +777     

Impacted Files	Coverage Δ
python/dask_cudf/dask_cudf/sorting.py	92.90% <0.00%> (-1.21%)	⬇️
python/cudf/cudf/io/csv.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/hdf.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/orc.py	0.00% <0.00%> (ø)
python/cudf/cudf/__init__.py	0.00% <0.00%> (ø)
python/cudf/cudf/_version.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/abc.py	0.00% <0.00%> (ø)
python/cudf/cudf/api/types.py	0.00% <0.00%> (ø)
python/cudf/cudf/io/dlpack.py	0.00% <0.00%> (ø)
python/cudf/cudf/core/frame.py	0.00% <0.00%> (ø)
... and 61 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 23de7d0...d6ebfb3. Read the comment docs.

[devavret]

Approving the changes in this PR.

However it's a bit disappointing that the aggregate functor model didn't work well with struct columns. And that we had to add mean, count, and covariance calculation to aggregate_result_functor::operator()<aggregation::CORRELATION> instead of just calling operator()<aggregation::COVARIANCE> and then use cache.get_result() to get the cov, stddev results out of it.

[karthikeyann]

we had to add mean, count, and covariance calculation to aggregate_result_functor::operator()aggregation::CORRELATION instead of just calling operator()aggregation::COVARIANCE

This is because of the "non-null values only" requirement for both covariance and correlation. If operator()<aggregation::COVARIANCE> is called, child columns mean & count will be recalculated for child columns with non-identical nulls.
Anyway it shares the same cache. So for identical nulls, it will still be able to use the cache and avoid the repeated computation.

I made changes to cache covariance if already calculated for this struct.

However it's a bit disappointing that the aggregate functor model didn't work well with struct columns

Since python requests correlation for pairs of columns as structs, caching for child column has added benefit of sharing std, mean, count for identical null columns across multiple aggregation requests. if we cache as struct column, this will not be possible.

[karthikeyann]

@gpucibot merge