Tight First-and Second-Order Regret Bounds for Adversarial Linear Bandits