[C++][Compute] Rank function considers NaNs and nulls equal

[pitrou]

Describe the bug, including details regarding any error messages, version, and platform.

Expected (NaNs and nulls are distinct, ordered according to null_placement):

>>> a = pa.array([1, None, math.nan, 2, math.nan, None])
>>> pc.rank(a, tiebreaker='min', null_placement='at_end')
<pyarrow.lib.UInt64Array object at 0x7f168295ca00>
[
  1,
  5,
  3,
  2,
  3,
  5
]
>>> pc.rank(a, tiebreaker='min', null_placement='at_start')
<pyarrow.lib.UInt64Array object at 0x7f1682845600>
[
  5,
  1,
  3,
  6,
  3,
  1
]

Actual (NaNs and nulls are considered ties):

>>> pc.rank(a, tiebreaker='min', null_placement='at_end')
<pyarrow.lib.UInt64Array object at 0x7f1682951660>
[
  1,
  3,
  3,
  2,
  3,
  3
]
>>> pc.rank(a, tiebreaker='min', null_placement='at_start')
<pyarrow.lib.UInt64Array object at 0x7f1682845600>
[
  5,
  1,
  1,
  6,
  1,
  1
]

Component(s)

C++

[timgrein]

Hey 👋

Recently started to look more deeply into Arrow, I would love to get my hands dirty with some smaller issues. Do you think this is one is a good starting point? If yes, I would like to give it a shot :)

[pitrou]

Hi @timgrein , yes, I think it would be a good issue to start if you already have some familiarity with the basic Arrow C++ APIs. However, there is a refactor going on in #45217 (probably soon merged), so you should probably base your work on that.

[pitrou]

Update: the aforementioned PR was merged, so you can start from git main :)