SimpleToM: Exposing the Gap between Explicit ToM Inference and Implicit ToM Application in LLMs