Multimodality for final LLM

[lighthea]

New Feature:

The final LLM answering to the query should take as input the images extracted from the files.

Specification:

• Multimodal LLMs supported
• Easy to extend for new architectures

[Stefanstud]

For the RAG part