Stream output to text editor

While jp is a CLI, it should support ergonomic workflows when chatting with the assistant.

One ideal case would be:

• jp query --new
• Open text editor for initial message
• On save, send message to LLM and wait for response
• Stream response back to editor, line-by-line, in same buffer
• Have some kind of "marker" to indicate end of response
• Wait for new data and for file to be saved again
• Parse file to get newest query and send together with history of conversation to LLM
• Wait for response, and stream back to editor again

This would also allow for more advanced use-cases (as a potential follow-up):

• In an ongoing conversation, removing part of the history rewrites that conversation, essentially allowing you to manipulate the entire state of the conversation.

Start by locating the CLI entry point for jp query --new and the existing assistant request and response flow. Define the editor interaction, response-end marker, save-and-resume behavior, and conversation-history parsing described here; the work is done when a saved query can receive a streamed response in the same buffer and be followed by another query.