Continuous state-space models for optimal sepsis treatment: a deep reinforcement learning approach