Code comments in processMemSetMemCpyDependence indicated that
the memset would be sunk below the memcpy when being rewritten. That
was however not true as the code actually inserted the MemSet before
the MemCpy. This patch updates the transform to actully insert the
MemSet after the MemCpy.
While being at it, this patch also make sure that the new memset is
reusing the debug location from the old memset rather than taking
the debug location of the memcpy.