- 2-3x faster on every size array that was benchmarked
 - breaking change: no longer retains the order of the original arrays.