[minor]: Update median implementation

[akurmustafa]

Which issue does this PR close?

Improves the situation in 13550

Rationale for this change

While examining the median code to see where can be changed to improve performance. I recognized that while calculating
the median for slices with even number of entries, we call select_nth_unstable_by twice. First to bipartition slice according to index n/2. Then to bipartition first part by index n/2-1. The second bipartitioning is indeed maximum calculation.

This PR changes the second bipartitioning call with maximum. I am not sure, whether this will improve the performance but I think at the worst case performance will be same, and this version is slightly more readable.

What changes are included in this PR?

Are these changes tested?

Existing tests

Are there any user-facing changes?

[Dandandan]

As unwrap is used, the code could be updated / simplified to work on arrays of size >= 1

[akurmustafa]

That makes sense to me, updated the implementation such that function doesn't return option in 0303c9

[Dandandan]

I think this should work

Suggested change

	array
	.iter()
	.fold(array[0], |max, &val| if max > val { max } else { val })
	}
	array
	.into_iter()
	.max()
	.unwrap()
	}

[akurmustafa]

When I did this I got following error:

^^^ the trait `std::cmp::Ord` is not implemented for `<T as arrow::array::ArrowPrimitiveType>::Native`, which is required by `&<T as arrow::array::ArrowPrimitiveType>::Native: std::cmp::Ord`

I guess, max expects Ord trait. However, T::Native implements PartialOrd trait.

[akurmustafa]

By the way, I could do this using max_by API.

[Dandandan]

Nice

[Dandandan]

I think we can use Ord though?

[akurmustafa]

I am not sure we can, at least I couldn't make it work. However, maybe I am missing something. By the way, following is the trait description for ArrowPrimitiveType::Native

pub trait ArrowNativeType:
    std::fmt::Debug + Send + Sync + Copy + PartialOrd + Default + private::Sealed + 'static
{

and signature of max is as follows:

    fn max(self) -> Option<Self::Item>
    where
        Self: Sized,
        Self::Item: Ord,
    {
        self.max_by(Ord::cmp)
    }

[comphead]

lgtm thanks @akurmustafa and @Dandandan for the review