INEFFICIENCY: We are storing 2 types of 128 token chunks in weaviate - one from SWR and one from AMR.

[Dhanushkmr]

• swr 128 tok chunks are done on the basis of page-by-page.
• amr 128 tok chunks are done recursively breaking down 2048 -> 512 -> 128

There are reasons to keep it this way because it helps with resolving chunk -> page for the frontend when we use SWR.

Possible solution: try to assign pages from chunks (amr 128 chunks) by looking at text overlap.