Two-sample rank tests for censored data with non-predictable weights