HugAgent: Benchmarking LLMs for Simulation of Individualized Human Reasoning