Think Smart, Act SMARL! Analyzing Probabilistic Logic Driven Safety in Multi-Agent Reinforcement Learning