Semantic Emoji Finder thanks to BERT support

[astrowonk]

I was able to use llama.cpp (via the python bindings) to generate BERT embeddings with relatively little memory (compared to pytorch and sentence_transformers) and allow for multi-word searches in my semantic emoji finder. I think a sentence-transformers version of this web app would use nearly 1GB of RAM; now implemented it's running in less than <200MB.

I've deployed the changes on the deployed Dash app.

[ggerganov]

Very cool!