Has it been settled whether robots.txt applies to user-driven chat sessions and if things like the crawl delay should be applied to say an end-user, an ip address, a harness provider, etc? My understanding is robots.txt is more for training exclusions, but less so for agent work.