Benchmarking Robustness of Contrastive Learning Models for Medical Image-Report Retrieval