PapersPlease: A Benchmark for Evaluating Motivational Values of Large Language Models Based on ERG Theory