Identifying Decision Points for Safe and Interpretable Reinforcement Learning in Hypotension Treatment