Self-play muTuAl Reasoning (rStar): A Novel AI Approach that Boosts Small Language Models SLMs’ Reasoning Capability during Inference without Fine-Tuning