Move trigger_ai_message_with_tool_call to conversation/generate.py

fix: remove formatting displayed message with references when no sources are used.

fix: clean up chunk formatting

feat: Adapt structured llm output to enable in-line citations & format final response with markdown containing citations.

feat: Structured LLM response for citation. Frontend rendering still broken.

fix: make model_provider call consistent in app_settings

feat: setup Amazon Bedrock for LLM (Claude Sonnet 3.5) and Embeddings (Cohere Embed multilingual v3)

Refactor loading of RAG prompt into function

Replace currently hard-coded prompt with loading production prompt from Langfuse

Load prompt from Langfuse

Update comments and only get langfuse handler instead of also the client

Initialize the Langfuse client in its own script and refactor code

Add Langfuse tracing via Langchain callback when generating answer

Basic set-up.