To avoid getting slammed with a worst-case input, we're going to pick our pivots randomly from the set of possible pivots.
Caveat: Our analysis will assume that elements in the list are distinct so there are no tie-breaking issues on the final position of the pivot. If we define the algorithm properly, duplicate elements in the list only help, so we ignore this case here.
