[FEA] Match pandas behavior for comparisons involving true nulls

[brandon-b-miller]

Is your feature request related to a problem? Please describe.
Nulls should propagate through comparisons the same way they do in Pandas. Any kind of comparison operation where <NA> is one of the operands should yield <NA>. This is not currently so in cuDF:

>>> cudf.Series([1,2,3]) > cudf.Series([1,2,None])
0    False
1    False
2    False
dtype: bool

What we should have is:

>>> pd.Series([1,2,pd.NA], dtype='Int64') < pd.NA
0    <NA>
1    <NA>
2    <NA>
dtype: boolean

Describe the solution you'd like
For all comparison operations, the mask of the resulting series should be the result of an elementwise logical or between the masks. I believe this is true in libcudf, and suspect we originally went a different way to match pandas nan logic instead of true null. The same thing should work with cudf.NA.

Describe alternatives you've considered

Additional context
xref : issue #7043

[jrhemstad]

suspect we originally went a different way to match pandas nan logic instead of true null.

I think this is the crux of it. Pandas until recently didn't have true nulls and everything was hacky using floats/NaN.

Not only does the proposed null logic make more sense, it's also higher performance because it doesn't require additional kernels/allocations to replace nulls w/ False.

[github-actions]

This issue has been marked stale due to no recent activity in the past 30d. 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 marked rotten if there is no activity in the next 60d.

[brandon-b-miller]

This is getting fixed this release (0.19)