Adjusting HIV prevalence estimates for non-participation: an application to demographic surveillance