ELABORATION: A Comprehensive Benchmark on Human-LLM Competitive Programming