Estimating the number of people who inject drugs using repeated respondent-driven sampling (RDS) in a community-based program: implications for the burden of hepatitis C and HIV infections and harm reduction coverage