Memorization of Named Entities in Fine-tuned BERT Models