Problem drug use prevalence estimation revisited: heterogeneity in capture–recapture and the role of external evidence