fayerman-source moves to stop the logs from spilling privileged material

fayerman-source spotted a quiet leak in how the fork's AI layer keeps records. Every raw exchange with the Claude model - the user's question, the document text fed in as context, and the model's answer - was being written, in full, to a plain text file on disk by default. Nobody rotates that file, nobody guards it, and in a legal workflow it is exactly the privileged material you do not want piling up unread on a server.

The proposed change deletes that automatic file and instead only prints the raw stream when an operator deliberately turns on a debug setting. Developers who need to inspect what the model is doing still can; a normal deployment writes nothing sensitive to disk. Worth noting: the change was filed against the main project but closed without being merged, so the default logging it targeted is still in place there.