Estimating HIV prevalence from surveys with low individual consent rates: annealing individual and pooled samples