Artifact comment threads can show the commenter's access level

What

• Comment and author data passed into the artifact-comments auto-reply feature now carries an access field (owner/editor/commenter/viewer) and an authoredByOthers flag, instead of only knowing that a message came from a human.
• A new phrase helper turns that into readable text like "the owner's", "an editor's", "a commenter's", or "a viewer's".
• When any comment in the visible thread has this access info, the prompt is told that a comment's header may show a tag such as [editor] or [owner, sent to you] in place of the word "human".
• The prompt also explicitly warns that if a comment's own text happens to contain a word like "owner", that's still just untrusted text typed by someone, not an actual permission grant.

Why

This lets Claude Code's auto-reply behavior take a commenter's real role into account (for example, treating an owner's request differently from an anonymous viewer's) while making sure it can't be fooled by someone typing a fake role claim into a comment.