Echoes of Socratic Doubt: Embracing Uncertainty in Calibrated Evidential Reinforcement Learning