On a singularity in the likelihood for a change-point hazard rate model