create AI policy

[bakkot]

We should have a policy on AI-authored PRs. I would like to accept them but require disclosure, and (preferably) to forbid using AI to write PR descriptions or commit messages.

[ptomato]

That matches my preference for a policy. I guess the PR descriptions and commit messages are already covered by TC39's policy.

As for disclosure I'd prefer discouraging Assisted-By: or Co-Authored-By: in favour of disclosure of the author's process — as a reviewer, I don't really care about which version of which model they used, but I do care about whether the tests were generated verbatim or hand-edited or whatever, and how they were verified.