Pull the HUMAN reviewer comments from a pull request and triage them into a worked
list: categorize each as SHOULD FIX, NITPICK, or QUESTION with a confidence rating,
ask about anything under 85% confidence, and save the list to the human-review state
file for /ai-next to work through. Use when the user says "pull the human comments",
"triage the reviewer feedback", "/hum-comments", or pastes a PR link asking to work
through what human reviewers said.