Realistic Evaluation of Deep Partial-Label Learning Algorithms