fix: pin PGDATA so the postgres 18 bump upgrades in place

postgres:18 (tracked by pgautoupgrade) relocated the default PGDATA from
/var/lib/postgresql/data to /var/lib/postgresql/18/docker. Since the db volume
is mounted at the legacy path, pg18 would init a fresh empty cluster there and
orphan all existing data (pgautoupgrade never runs the upgrade). Pinning PGDATA
to the legacy path makes pg18 find the existing cluster and upgrade it in place.

compose.mas.yml

@@ -5,7 +5,7 @@ version: "3.8"
 
 services:
   mas:
-    image: ghcr.io/element-hq/matrix-authentication-service:1.17.0
+    image: ghcr.io/element-hq/matrix-authentication-service:1.18.0
     command: ["server", "--config=/etc/mas/config.yaml"]
     environment:
       - DOMAIN

compose.yml

@@ -43,7 +43,7 @@ services:
       start_period: 2m
 
   app:
-    image: "matrixdotorg/synapse:v1.153.0"
+    image: "matrixdotorg/synapse:v1.154.0"
     volumes:
       - "data:/data"
     secrets:
@@ -114,7 +114,7 @@ services:
       restart_policy:
         condition: on-failure
       labels:
-        - "coop-cloud.${STACK_NAME}.version=7.2.0+v1.153.0"
+        - "coop-cloud.${STACK_NAME}.version=7.3.0+v1.154.0"
         - "coop-cloud.${STACK_NAME}.timeout=${TIMEOUT}"
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:8008/health"]
@@ -128,6 +128,10 @@ services:
     secrets:
       - db_password
     environment:
+      # postgres:18 relocated the default PGDATA to /var/lib/postgresql/18/docker.
+      # Pin it to the legacy path so pgautoupgrade finds the existing cluster on the
+      # mounted volume and upgrades it in place, instead of init'ing an empty one.
+      - PGDATA=/var/lib/postgresql/data
       - LC_COLLATE=C
       - LC_CTYPE=C
       - POSTGRES_DB=synapse