A Bayesian model for network flow data: an application to BikeMi trips