[C++] Add variadic row-wise min/max kernels (least/greatest)

[asfimport]

Add a pair of variadic functions equivalent to SQL's least/greatest or R's pmin/pmax. Should take 0, 1, 2, ... same-length numeric arrays as input and return an array giving the minimum/maximum of the values found in each position of the input arrays. For example, in the case of these 2 input arrays:

Array<double>        Array<double>
[                    [
  1,                   2,
  4                    3
]                    ]

least would return:

Array<double>
[ 
  1,
  3
] 

and greatest would return

Array<double>
[ 
  2,
  4
] 

The returned array should have the same data type as the input arrays, or follow promotion rules if the numeric types of the input arrays differ.

Should also accept scalar numeric inputs and recycle their values.

Reporter: Ian Cook / @ianmcook
Assignee: David Li / @lidavidm

Related issues:

• [C++] Arithmetic kernels for numeric arrays (is a child of)
• [R] Add bindings for pmin() and pmax() (is depended upon by)

PRs and other links:

• GitHub Pull Request #10390

_{Note: This issue was originally created as ARROW-12751. Please see the migration documentation for further details.}

[asfimport]

Joris Van den Bossche / @jorisvandenbossche:
Numpy has this as np.minimum and np.maximum, although those are limited to a fixed number of 2 input arrays (so a binary min/max, not variadic)

[asfimport]

David Li / @lidavidm:
Just to clarify, do we want to propagate nulls or ignore them? (NumPy raises an error which is also an option)

minimum([1, 2, 3], [null, 1, 1]) = [null, 1, 1] # Propagate
minimum([1, 2, 3], Scalar(null)) = [null, null, null]
minimum([1, 2, 3], [null, 1, 1]) = [1, 1, 1]    # Ignore
minimum([1, 2, 3], Scalar(null)) = [1, 2, 3]

Also, what do we do with 0 arguments?

minimum(Scalar(null)) = null or error?
minimum() = [], NullScalar, or error? 

[asfimport]

Ian Cook / @ianmcook:
In principle this should behave the same as the min/max aggregate kernels, I think.

Can we use ScalarAggregateOptions or something like it to control whether to propagate vs. ignore?

With zero arguments it should return null.

[asfimport]

Ben Kietzman / @bkietz:
Issue resolved by pull request 10390
#10390