[RAG] Fix RAG Passage Loading

[klshuster]

Patch description
Previously, if we errored out attempting to read a .csv file, we switched to reading passages and splitting by \t manually. However, we were still referencing an old variable from the prior loop, which references an old row. I've updated to point to the new variable.

Testing steps
I discovered this because someone pointed out that their retriever was retrieving the same document for every turn. I looked at the logs and saw that the indexing for 5k documents was taking over 5 minutes; this is way too slow, and usually means that the passage embeddings are too close to eachother. After reproducing the error locally, the passages are correctly loaded and indexing takes around 2 seconds.

Relying on CI for the rest.