chore(deps): postgres 18, and the data directory it now insists on

.agents/skills/infra/docker-dev/SKILL.md

@@ -47,6 +47,12 @@ docker compose -f docker/compose.yaml --profile postgres up
 docker compose -f docker/compose.yaml --profile minio up      # console on 9001
 ```
 
+If `postgres` exits 1 before printing a single server line, and the message names
+`pg_ctlcluster` and a major-version directory, the volume was written by an older major
+version. Postgres 18 keeps its cluster under `/var/lib/postgresql/18/` and will not adopt
+one left by 16. `docker volume rm visionset_postgres-data` clears it; nothing reads this
+service, so there is nothing in there to keep.
+
 ## The three ways to run it
 
 The stack has three permanent, mutually compatible configurations. They differ in one

docker/compose.yaml

@@ -265,15 +265,22 @@ services:
   # why they sit behind profiles and are off by default.
   postgres:
     profiles: ["postgres"]
-    image: postgres:16
+    image: postgres:18
     environment:
       POSTGRES_DB: visionset
       POSTGRES_USER: visionset
       POSTGRES_PASSWORD: visionset
     ports:
       - "5432:5432"
     volumes:
-      - postgres-data:/var/lib/postgresql/data
+      # The whole directory, not the `data/` beneath it. From 18 the image keeps its
+      # cluster in a major-version subdirectory and refuses to start at all — exit 1
+      # before any server — when it finds a mount on the old path, so bumping the tag
+      # without moving this line takes the profile down. A cluster written by 16 is
+      # unreadable to 18 wherever it is mounted: `docker volume rm
+      # visionset_postgres-data` is the way through, and costs nothing while nothing
+      # is wired to this service.
+      - postgres-data:/var/lib/postgresql
 
   minio:
     profiles: ["minio"]