fix(deploy): health-check via docker network, not host port

The VPS doesn't publish the app port to the host (Traefik reaches it over
the docker network), so curl http://localhost:PORT/health always failed —
a false negative that could trigger an unwanted automatic rollback on a
successful deploy. Check /health from INSIDE the container via
`docker compose exec` instead.

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

scripts/deploy-vps.sh

@@ -42,7 +42,17 @@ if [[ -z "${PORT:-}" ]]; then
     fi
 fi
 PORT="${PORT:-9000}"
-HEALTH_URL="http://localhost:${PORT}/health"
+# Health verificato DENTRO la rete del container: in produzione la porta NON è
+# pubblicata sull'host (Traefik la raggiunge via rete docker), quindi un curl da
+# localhost darebbe un falso negativo (e rischierebbe un rollback inutile).
+HEALTH_URL="http://127.0.0.1:${PORT}/health"
+
+# 0 se /health risponde 200 dall'interno del container.
+app_health() {
+    docker compose exec -T "$SERVICE" python -c \
+"import urllib.request,sys; sys.exit(0 if urllib.request.urlopen('${HEALTH_URL}').status==200 else 1)" \
+        >/dev/null 2>&1
+}
 
 # ─── Pre-check ───────────────────────────────────────────────────────────
 command -v git >/dev/null    || { echo "FATAL: git non installato"; exit 1; }
@@ -125,10 +135,11 @@ docker compose up -d
 echo "==> attendo /health (timeout ${HEALTH_TIMEOUT_SECONDS}s, retry ogni ${HEALTH_INTERVAL}s)"
 deadline=$(( $(date +%s) + HEALTH_TIMEOUT_SECONDS ))
 while [[ $(date +%s) -lt $deadline ]]; do
-    if curl -fsS "$HEALTH_URL" >/dev/null 2>&1; then
+    if app_health; then
         echo
         echo "==> health OK"
-        curl -s "$HEALTH_URL"
+        docker compose exec -T "$SERVICE" python -c \
+"import urllib.request; print(urllib.request.urlopen('${HEALTH_URL}').read().decode())" 2>/dev/null || true
         echo
         echo
         echo "==> deploy DONE (SHA $CURRENT_SHA → $NEW_SHA, branch $BRANCH)"