Chatbot improvements

[miguelgrinberg]

This PR includes a set of minor improvements and tunings to the RAG chatbot example:

• The condensed question is used only to retrieve relevant documents from the ES index. Then the LLM receives the original chat history and question from the user with this context. This prevents user questions from diluting as more follow up questions are entered.
• A longer request timeout was added to the bulk importer, for robustness.
• The source parser used in the frontend was expanded to handle a common case in which the LLM does not select any sources from the provided context.