feat(ai-help): index text-embedding-3-model embeddings

[caugner]

Summary

Problem

We want to upgrade our embeddings to the text-embedding-3-model model.

Solution

Index all documents in parallel using both the current and the new model.

---

How did you test this change?

Ran yarn ai-help-macros update-index locally, and it populated the embedding_next column in the Supabase dev project (the column also exists in the stage/prod projects).

[fiji-flo]

Conceptual feedback:

While there is no next we should set EMBEDDING_MODEL_NEXT to null. In that case an existing document has embedding_next set we should also set it to null.

So when a model stabilized we can move from embedding_next to embedding and then set EMBEDDING_MODEL_NEXT to null.

[caugner]

While there is no next we should set EMBEDDING_MODEL_NEXT to null. In that case an existing document has embedding_next set we should also set it to null.

So when a model stabilized we can move from embedding_next to embedding and then set EMBEDDING_MODEL_NEXT to null.

Sure, fixed in d8177d3.

[fiji-flo]

My main point as per our slack conversation still stands. This is not necessary to evaluate the new model and ads complexity to our code. Since the PR is written and you keep updating it, even after me pointing out that we should just evaluate it fully locally first, there's no point blocking it because time has already been spend writing and reviewing.

It will be more hacky to test embedding dimensions and we don't yet know which embeddings we want to launch with but I won't block this as it eats up time.