This phase add counters for the dynamically executed number of nodes. Incrementing the counter
for each node would be too costly, so this phase takes the compromise that it trusts split
probabilities, but not loop frequencies. This means that it will insert counters at the start of
a method and at each loop header.
A schedule is created so that floating nodes can also be taken into account. The weight of a node
is determined heuristically in the
ProfileCompiledMethodsPhase.getNodeWeight(Node)
method.
Additionally, there's a second counter that's only increased for code sections without invokes.