Safeguarded Progress in Reinforcement Learning: Safe Bayesian Exploration for Control Policy Synthesis