ProvablyEfficientReinforcementLearningwith LinearFunctionApproximationunderAdaptivity Constraints