Why do we use quadratic variation in stochastic calculus in the ito lemma?