The Evolving Landscape of LLM- and VLM-Integrated Reinforcement Learning