Understanding the declining Canadian homicide rate: a test of Holinger's relative cohort size hypothesis