AI policy update: gather lessons over the last year and iterate

We did a big RFC and came up with our AI policy: https://llvm.org/docs/AIToolPolicy.html

Things change fast, and I'm sure there are improvements we can make.

Right off the bat, we should scrap the extractive labelling, and replace it with github moderation. The tools are there, we should just use them. I admit, I overthought things.

With regard to labelling AI-assisted contributions, we didn't specify any standard. Personally, I feel like Assisted-by: some LLM brand is free product placement in our commit logs, and I don't like that. OTOH, people have been building ad-hoc dashboards measuring AI usage by scraping the tags. The original labelling motivation was mostly about communicating to human reviewers the PR drafting methodology, to set their expectations about how much human thought went into each code token. I copied the Fedora AI policy text about sharing best practices without thinking critically about it.

I'm also inclined at this point to make a sharper policy distinction between spaces intended for human communication and code. I have little patience for Claude-speak RFCs, or issue comments, but I'm willing to stamp a big refactoring to non-essential Python lit code if I can review the testing methodology. There's room for using LLMs as editing aides, but I think we could expand scope of the "write your own commits" policy to something like "do not use LLMs to draft Github comments, Discourse posts, or PR descriptions, only use them as editors".