A Framework for History-Aware Hyperparameter Optimisation in Reinforcement Learning