Follow-the-Perturbed-Leader with Fr\'{e}chet-type Tail Distributions: Optimality in Adversarial Bandits and Best-of-Both-Worlds