A Provably Efficient Option-Based Algorithm for both High-Level and Low-Level Learning