fix(docker): remove redundant custom network from prod compose

nginx sat on two Docker networks (Coolify's shared per-app network,
where coolify-proxy lives, plus our custom "mandant" bridge). With
no explicit traefik.docker.network label, Traefik's Docker provider
picked one non-deterministically across deploys — when it picked
"mandant", its backend connection went to an unreachable network and
silently hung after completing the TLS/HTTP handshake with clients.

All services already share Coolify's injected network, so the custom
network was redundant. Removing it leaves each container on exactly
one network, eliminating the ambiguity.

docker/docker-compose.prod.yml

@@ -16,8 +16,6 @@ services:
     depends_on:
       mysql:
         condition: service_healthy
-    networks:
-      - mandant
 
   nginx:
     build:
@@ -27,8 +25,6 @@ services:
       - "8899:80"
     depends_on:
       - backend
-    networks:
-      - mandant
 
   mysql:
     image: mysql:8.0
@@ -44,12 +40,6 @@ services:
       interval: 5s
       timeout: 5s
       retries: 10
-    networks:
-      - mandant
 
 volumes:
   mysql_data:
-
-networks:
-  mandant:
-    driver: bridge