On Event Individuation for Document-Level Information Extraction