The two-sample problem for failure rates depending on a continuous mark: an application to vaccine efficacy