assertions on ai output, not vibes

Cursor wrote an eval that checks whether the model mentioned Paris somewhere in its answer. It used exact equality, which fails because the model returned a full sentence. Switch the check to a contains comparison so it passes.

Expected output:

pass