Optimal Analysis for Bandit Learning in Matching Markets with Serial Dictatorship