SwS: Self-aware Weakness-driven Problem Synthesis in Reinforcement Learning for LLM Reasoning