Generating Visual Stories with Grounded and Coreferent Characters