Implement a different chain to ensure the LLM doesn't miss important data in chunks not passed to it.

Change `create_stuff_documents_chain()`

[Official Docs](https://python.langchain.com/docs/modules/chains/)

Look at the existing `create_stuff_documents_chain()` function in the codebase to understand its current implementation. Review LangChain documentation on different chain types (like map-reduce, refine) to see alternatives that might handle chunk data more comprehensively. Determine what 'important data in chunks not passed to it' means in this context—likely related to retrieval or chunk processing. The goal is to modify or replace the chain to improve data coverage.

索引モデルが issue の本文から書いたものです。