fix: Fix performance regression for sort/gather on list/array columns

[nameexhaustion]

Fixes #19559

It turns out, our slow-paths for take_unchecked are very slow, because the implementation constructs a Growable with a Vec containing 1 unit-length Array per gather index

polars/crates/polars-arrow/src/compute/take/list.rs

Lines 35 to 49 in ebeeea7

	let arrays = indices
	.iter()
	.flat_map(|opt_idx| {
	opt_idx.map(|index| {
	let index = index.to_usize();
	let slice = values.clone().sliced(index, 1);
	capacity += slice.len();
	slice
	})
	})
	.collect::<Vec<ListArray<I>>>();
	let arrays = arrays.iter().collect();
	if let Some(validity) = indices.validity() {
	let mut growable: GrowableList<I> = GrowableList::new(arrays, true, capacity);

And all of them end up being bit-counted in Growable::new :P

polars/crates/polars-arrow/src/array/growable/list.rs

Line 49 in ebeeea7

if !use_validity & arrays.iter().any(|array| array.null_count() > 0) {

Also, the reason why we regressed in list-gather performance was because we were previously not using that super-slow list gather implementation until a dispatch impl ChunkTakeUnchecked<IdxCa> for ListChunked { was added to it in #19327

[ritchie46]

Thanks Simon. That slow path was the reason I made the change. I didn't realize we did never hit it though.