RAG implementation

[satyamshukla20]

This PR introduces Retrieval-Augmented Generation (RAG) into the presentation generator by integrating pgvector for embeddings and contextual retrieval. Uploaded files are processed into chunks and stored in the document_chunk table, organized by organization and user-selected tags to enable efficient and meaningful retrieval. When creating a presentation, the LLM now leverages this stored knowledge as context, resulting in more relevant and accurate outputs tailored to the user’s documents. Users can also view all presentations they have created, ensuring a seamless workflow from document upload to contextualized presentation generation. This enhancement significantly improves the quality, personalization, and usability of generated presentations.