Add support for multilingual embedding models

[hellojwilde]

EnergeticAI currently uses the lite Universal Sentence Encoder model, which is English only: https://tfhub.dev/google/universal-sentence-encoder-lite/2

It's possible we could convert one of the smaller multilingual models on TFHub and explore if there's a way to get it to be usable: https://tfhub.dev/google/universal-sentence-encoder-multilingual/3