Towards Reinforcement Learning for Exploration of Speculative Execution Vulnerabilities