Omni-AutoThink: Adaptive Multimodal Reasoning via Reinforcement Learning