### What problem does this PR solve?
This PR organize chunks in the prompt by document and indicate what is
the name of the document in this way
```
Document: {doc_name} \nContains the following relevant fragments:
chunk1
chunk2
chunk3
Document: {doc_name} \nContains the following relevant fragments:
chunk4
chunk5
```
Maybe can be a baseline to add metadata to the documents.
This allow in my case to improve llm context about the orgin of the
information.
### Type of change
- [X] New Feature (non-breaking change which adds functionality)
Co-authored-by: Miguel <your-noreply-github-email>
41a0601735