* cli : persist reasoning_content in chat history llama-cli collected reasoning from the stream for display but only stored assistant content in messages, so --reasoning-preserve could not re-inject prior thoughts on later turns.
