Switch to Whoosh implementation to improve speed

[eepMoody]

Replaces the existing hand-tuned search with the Whoosh framework.

This has major performance benefits, but incurs a much higher indexing time. To deal with this, I've adopted a strategy that caches indexes in the repo, unpacks them for quick development purposes, then runs a background indexing process inside the server after deployment. In practice, this means search results may start slightly out-of-date (if a major content release has not been cachced) but will be consistent within approximately 5 minutes of deploy.

Major changes:

• Replace hand-tuned and ES options with Whoosh
• replace vector search with spaCy embeddings for true semantic matching
• add management commands, including quickindex which will index and create zipped files for checking in
• add tests covering all search methods, ensuring they return realistic results
• add background process in server to index live content in production

NOTE being kept in draft until #844 is merged, as this is based on that branch

[augustjohnson]

@eepMoody what's the status on this PR? If I understand correctly, a fair amount of the search is already in place in staging... Where does this fit in?

[eepMoody]

Ah sorry, I hit a bit of a wall on getting the actual switch to Elasticsearch (which is a major performance boost) sorted out. What's in production adds the new search methods, but the performance isn't really improved. Might be a little bit worse, but hard to tell until it hits real resources.

The main bit that's outstanding is the deployment setup, since ES needs to run in its own container/server. I haven't quite gotten that into a state where it's functional in production.

If you want to take a peek, maybe there's something obvious that I've overlooked or another approach that might make more sense?