Add guidance: reviewers of AI work should try to invalidate it

[d-morrison]

Migrated from UCD-SERG/lab-manual#314 (the Working-with-AI content now lives in this repo).

Original request (d-morrison):

as reviewers of AI work, our job is to invalidate the AI's work

we should search mercilessly for mistakes and imperfections; put on your harshest critic hat.

Suggested home: the pull-request-workflow / reviewing chapter — a short section on the reviewer's stance toward AI-generated output: assume it's wrong until verified, actively hunt for hallucinated citations/functions/URLs, and treat plausible-sounding prose as the main risk rather than a comfort.

[github-actions]

A GitHub issue proposes adding guidance for reviewers of AI-generated work, emphasizing a critical stance during review. It suggests that reviewers should actively try to invalidate the AI's output by rigorously searching for errors, hallucinated citations, fabricated data, or plausible-sounding yet inaccurate prose. The recommended placement for this guidance is within the pull-request workflow or reviewing chapter, highlighting the importance of assuming AI outputs are incorrect until thoroughly verified. This issue stems from feedback migrated from another repository.