Bias and Sensitivity Analysis when Estimating Treatment Effects from the Cox Model with Omitted Covariates