Duration dependence versus unobserved heterogeneity in treatment effects: Swedish labor market training and the transition rate to employment