What’s so special about BERT’s layers? A closer look at the NLP pipeline in monolingual and multilingual models