Investigate whether we could utilize different handler for sampling address and collecting hardware events

[tongping]

Now the overhead for reverse_index is around 3.5X and the overhead for canneal is aroudn 60%???

We could use different handler and simplify the logic.