The promise and limits of LLMs in constructing proofs and hints for logic problems in intelligent tutoring systems