[2409.16824] Uncertainty Representations in State-Space Layers for Deep Reinforcement Learning under Partial Observability