ComputationallyEfficientHorizon-Free ReinforcementLearningforLinearMixtureMDPs