bug(api) RAG uses first message in list instead of last

[alekst23]

Description

Our current implementation of RAG performs a search using the first message in the thread, which will always be the same. The search should instead be performed with one or more of the last messages from the thread.

Steps to reproduce

To reproduce the problem you will need 5 "hay" files and 1 "needle". You will also need a thread with two messages: first message about hay, second message about needle.

A RAG query will return 5 documents. Therefore, if the search is performed using the first message about hay, the latest user query about the needle will not be answerable because no matching documents will be returned.

Solution

We should be searching RAG using the last message in the thread