ReMiT: RL-Guided Mid-Training for Iterative LLM Evolution | ResearchPod