Text this: Unpacking Performance Variability in Deep Reinforcement Learning: The Role of Observation Space Divergence