Bottom-up heapsort and fix sort test.

Bottom-up heapsort and fix sort test.

[bungeman, 2013-01-24 08:06:25]

This change to heapsort to make it a 'bottom-up' heapsort improves asymptotic
performance from a C of 2 to a C of 1.5. In all benches but the 'repeated' bench
this shows some slight and predictable improvement. In the 'random' bench the
improvement is ~5%. This heapsort is still takes 1.3x time of skqsort in the
'random' bench.

The change to the sort test was (and is) needed to verify that any sort changes
actually work. The change here is to compare sksort output against the output of
qsort, as it is assumed that the system qsort is correct (or at least defines
the system sorting).

[reed1, 2013-01-24 13:13:15]

lets land the unittest upgrade first

[bungeman, 2013-01-24 22:13:39]

Now slightly faster and plays well with size_t.

[reed1, 2013-01-24 22:16:36]

lgtm

[bungeman, 2013-01-24 23:01:29]

Committed revision 7382.