Evaluating LLM-Contaminated Crowdsourcing Data Without Ground Truth