Performance of predecessors, successors, and CFG computation

[willtebbutt]

At present, the performance of compute_all_predecessors, compute_all_successors, and control_flow_graph are not bottlenecking anything (that I'm aware of), but there are clear redundancies in their implementations. For example:

• compute_all_predecessors makes use of compute_all_successors in its implementation. Often these are both needed at the same time, which means that compute_all_successors is often run twice.
• Both compute_all_predecessors and compute_all_successors create more allocations than is strictly necessary
• Type-stability hasn't been checked
• control_flow_graph makes use of both of these functions, and also has not really been optimised.

All this is to say that there is probably quite a lot that we could do to improve the performance of these functions, given some time.

If I ever notice that these are becoming an actual issue, I will remove the "low priority" tag, and prioritise solving this. In the mean time, this is probably a good place for someone to start working on the codebase.