BrokenMath: A Benchmark for Sycophancy in Theorem Proving with LLMs