feat(chat): Fix LLM RAG workflow issue to use RAG for nuanced or rephrased queries.

[oindrila-b]

Currently, the RAG workflow in our app has an issue where, if a user asks a question about a specific topic and then rephrases the question, the app relies solely on conversation history rather than using RAG to provide a more accurate response. As a result, the app can get "stuck in a loop," repeatedly pulling answers from previous messages without re-engaging RAG — even when the question's wording changes. This pattern continues until a query about an entirely different topic is introduced.

The logs reflect this behavior with the following message:
We were able to find the answer/respond to user's query in the conversation itself so not applying RAG.

To address this, we want to adjust the logic as follows:

Use conversation history for simple or contextual queries: The app should continue using the conversation history for straightforward interactions, such as greetings (e.g., "Hi") or basic fact-checking where the answer is already in context.

Use RAG for rephrased queries: When users ask slightly more complex questions — especially when they revisit a topic with different wording — the app should prioritise RAG over relying solely on past messages. This ensures we deliver the most relevant and accurate information without getting stuck in a conversational loop.

While this is an immediate fix, we plan to implement a more refined approach in the future to better balance the use of conversation history and RAG.