Evaluating LLM-Generated Obfuscated XSS Payloads for Machine Learning-Based Detection