Accelerated and instance-optimal policy evaluation with linear function approximation