Collect statistics that help understand Prepack effects

Prepack runs the global code, and we could collect some runtime information to help assess what Prepack does:

• How many objects get created when running the interpreter? (And how many survive and get serialized?)
• How many object properties get updated? (And how many are serialized?)
• How many times does each type of AST node get evaluated? (Pretty much all of those computations, except for some literals, object allocations, property updates, will get eliminated by Prepack.)

When forking and merging control-flow, compute the maximum value for each explored case.