This change allows us to avoid O(n^2) performance when partially sorted data
is given to NS_QuickSort. We no longer attempt an insertion sort if the input
appears pre-sorted. This brings the expected performance back to O(n*log(n))
but also eliminates the possibility of O(n) best case performance. Attempts
to achieve O(n) performance should be the responsibility of callers as they
are in a better position to evaluate the costs/benefit trade-off of looking
for special cases that can be sorted quicker.
