Gravar-mail: Empirical-likelihood-based semiparametric inference for the treatment effect in the two-sample problem with censoring