Which pivot selection strategy in Quicksort is usually the most effective at avoiding the $O(n^2)$ worst-case runtime across arbitrary input distributions?