fix(safety): minor flag wording -> child markers + adult-exclusion (Variant B)

A/B on the real Qwen guard showed the "young-looking" wording fixed the
in-arms-infant false negative but introduced a false positive on a youthful
adult. Variant B (explicit child markers + "an adult who merely looks
youthful/petite is FALSE") keeps the infant fix with zero false positives
(FN 1, FP 0 on the 10-image test set). Remaining FN is a tiny background
child the Qwen-4B guard can't resolve — a model limit, not a wording one.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

app.py

@@ -558,7 +558,7 @@ Look carefully at what is ACTUALLY visible in the image, then output ONLY this J
   "rating": "<0_SAFE | 1_MILD_SUGGESTIVE | 2_STRONG_SUGGESTIVE | 3_NON_EXPLICIT_NUDITY | 4_EXPLICIT_SEXUAL | 5_PROHIBITED>",
   "confidence": 0.0,
   "flags": {
-    "minor_or_young_looking": "<is any minor or young-looking human subject anywhere in the image — including small, distant, background, blurred, partial, or secondary people, not just the main subject? true / false>",
+    "minor_or_young_looking": "<TRUE if ANY person anywhere in the image — including small, distant, background, blurred, partial, or secondary people — appears to be a CHILD / under 18 (infant, toddler, child, or clearly pre-teen/teen face and body). An adult who merely looks youthful or petite is FALSE. else false>",
     "sexualized_minor_risk": false,
     "nudity": false,
     "explicit_genitals": false,