You're right, I too was a bit disappointed that the page didn't mention contextual embeddings since that has been the best performing and most robust approach in NLP for nearly four years now.

It's still a clear introduction to word vectors though and it is good to know where we came from.