LoopTune: Optimizing Tensor Computations with Reinforcement Learning