An Empirical Study Towards Prompt-Tuning for Graph Contrastive Pre-Training in Recommendations