Benchmarking Debiasing Methods for LLM-based Parameter Estimates