Improved Algorithm for Adversarial Linear Mixture MDPs with Bandit Feedback and Unknown Transition