On the issue of homophily in respondent-driven sampling: notes based on the case of men who have sex with men in Belo Horizonte, Brazil