Inconsistent default values for null equality and NaN equality

[bdice]

In cudf::distinct, we default to equal nulls and equal NaNs. This aligns with the defaults of intersect_distinct, union_distinct, and difference_distinct in PR #11043 (as of this writing - not yet merged).

However, there are competing standards:

• In make_collect_set_aggregation, we default to equal nulls and unequal NaNs.
• In drop_list_duplicates, we default to equal nulls and unequal NaNs.

We should strive for a consistent set of defaults for how null and NaN equality are handled. I think I would recommend a change in make_collect_set_aggregation and drop_list_duplicates to align with the other functions mentioned above.

Originally posted by @bdice in #11043 (comment)

[bdice]

Partially resolved by #11236, which removes drop_list_duplicates.

[github-actions]

This issue has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be labeled inactive-90d if there is no activity in the next 60 days.

[bdice]

I caught another place where nan_equality::UNEQUAL is being used in make_merge_sets_aggregation.

cudf/cpp/include/cudf/aggregation.hpp

Line 592 in 536ddd0

nan_equality nans_equal = nan_equality::UNEQUAL);

I will make a follow-up PR that can be merged after #11621.

[ttnghia]

How about renaming nan_equality::ALL_EQUAL to just EQUAL to match with null_equality?

[vyasr]

Not quite done yet. @bdice be careful with the wording on your PRs :) GH sees "partially resolves X" as ".... resolves X" and still closes the issue AFAICT