Previously I counted comparisons when there were no comparisons.

* incremental.rs: A call to `add_edge` was placed at the wrong place.

Now the correct solution, as suggested by the professor, is
implemented as the function `smi`.  :D

Now the double for loop is eliminated and the time complexity is
indeed O(n log(n)).

Also added (micro)-benchmarks to roughly confirm the time complexity
is as predicted.

The sorting function is complete, and accepts any type that has a
partial order.