Building a Japanese Document-Level Relation Extraction Dataset Assisted by Cross-Lingual Transfer