fix(stack): bring-up fixes from first local deployment

- traefik pin v3.3 -> v3.6: v3.3's old docker client fails against
  Engine 29 socket proxy (empty daemon errors, no routers)
- backend/minio/chatbot/umami: traefik.docker.network hint — Traefik
  picked the internal-network IP for multi-network services and replies
  were hairpin-dropped (all backend routes hung)
- db-init: psql needs an explicit -d arameshplaza (no default db 'aramesh'),
  set -e so failures actually fail; one-shot folded-YAML needs semicolons
- chatbot image: drop g++ (apt unreachable in build network; all wheels
  prebuilt) and mkdir /data before chown (VOLUME does not create it)
- frontend image: create caddy user (alpine image has none) with writable
  /config and /data
- monitoring: blackbox probes replace raw metric scrapes of JSON/HTML
  health endpoints (they reported 'down' while healthy); blackbox joins
  edge; SITE_URL passed via env (Prometheus has no ${VAR:-default}
  expansion); ContainerMemoryNearLimit PromQL fixed (comparison inside
  label braces); HostDiskAlmostFull filters WSL/Docker-Desktop
  pseudo-mounts
- alertmanager: config rendered by stack.sh (no env expansion in AM) and
  delivered to Bale (tapi.bale.ai); stack.sh monitoring command
- machine-local override: minio console/API moved to 29001/29000

docker-compose.monitoring.yml

@@ -27,6 +27,9 @@ services:
       - ./monitoring/prometheus:/etc/prometheus:ro
       - prometheus_data:/prometheus
     networks: [monitoring, edge]
+    environment:
+      # پروب عمومی — در تولید SITE_URL در .env تنظیم می‌شود (مثل https://aramesh.example)
+      SITE_URL: ${SITE_URL:-http://frontend/}
     healthcheck:
       test: ["CMD", "wget", "-q", "--spider", "http://localhost:9090/-/healthy"]
       interval: 30s
@@ -108,7 +111,8 @@ services:
     image: prom/blackbox-exporter:v0.25.0
     restart: unless-stopped
     logging: *default-logging
-    networks: [monitoring]
+    # edge برای رسیدن به سرویس‌هایی که پروب می‌شوند (frontend/minio/umami/chatbot/backend)
+    networks: [monitoring, edge]
     deploy:
       resources:
         limits: