Add a qsort implementation to Mibench/qsort

[nickwanninger]

This PR adds an implementation of quicksort to the MiBench patches, as calling the libc implementation of quicksort isn't super interesting. I've inlined a version of qsort that DSWP is used, but gets no speedup (the non-expanded input_large.dat takes 3 seconds baseline while the noelle version takes 48 seconds). I am working on an enabler to allow llvm's tailcallelim pass to perform it's work w/o giving up.

The main reason llvm is giving up seems to be that they expect tailcalls to look like this:

   call qsort( ... )
   call qsort( ... )
   ret

where the return is immediately after a recursive call to the function. However, the frontend spits out this code:

another_block:
   br cleanup

recursive_block
   call qsort( ... )
   call qsort( ... )
   call llvm.lifetime.end( ... )
   br cleanup
cleanup:
   ret

and as such, the tailcallelim pass gives up.