Multi-agent LLM workflows that propose, critique, test, debug, and refine algorithms through structured collaboration