From N-grams to Pre-trained Multilingual Models For Language Identification