Did the Models Understand Documents? Benchmarking Models for Language Understanding in Document-Level Relation Extraction