Improve logic used for sorting top-level expressions

The current sorting is done by type of expression and then by size of the "construct". It seems this is done before the naming unification because it can decrease the quality of the representation. For bird watcher, the order of functions/tasks is different from representation to representation although it is the same in the solution code. That reduces the likelihood that two solutions have the same representations.

Commenting out the sorting let to somewhat better results (no massive improvement though). In general, the sorting is a good thing though but it seems we should fine tune the logic so that it keeps the order of the tasks. That also makes it easier to reconcile the representation with the student code when looking at it.