Use BM25 for simple search

[lefnire]

The search bar currently does the following:

1. If the user typed < n_words (3 currently I think?), do a literal search (entries.filter(e => e.texts.toLowerCase().includes(search.toLowerCase())
2. If > n_words, do a cosine similarity search using sentence_transformers.semantic_search.

I was previously using Haystack for this, which has a few conditionals which would make this ticket easier. But the project was a lot beefier for too little payoff than I anticipated, so I just went back to the basics. So, if we want a more sophisticated search pipline, we can consider:

1. Try Haystack or Jina again. I'd lean towards Jina, they're making strides.
2. Consider using a Vector Database. This would be a re-architecting of the current system which uses PyArrow with vectors stored on S3 as parquet files. My solution is infinitely scalable and cheap. But a true Vector Database would add a lot of utility, like BM25 searching, ML utilities like sentiment analysis, question-answering, etc. It would require a Docker setup, and maintenance though.