fix: mount the scaffolded database volume where PostgreSQL 18 stores data

[Bccorb]

Fixes a defect introduced by #164, caught while doing the same bump in seamless-templates (fells-code/seamless-templates#42).

The bug

#164 moved the image tag but not the volume mount. PostgreSQL 18+ images store data in a major-versioned subdirectory (/var/lib/postgresql/18/docker), so a mount at /var/lib/postgresql/data is ignored and the container refuses to start:

Error: in 18+, these Docker images are configured to store database data in a
       format which is compatible with "pg_ctlcluster" ...
       Counter to that, there appears to be PostgreSQL data in:
         /var/lib/postgresql/data (unused mount/volume)

It restart-loops. A project scaffolded from that change had a database that never came up. See docker-library/postgres#1259.

Why CI missed it, and why that matters

The verify / verify job passed on #164 and I reported it as end-to-end proof. It wasn't. The harness's postgres service mounts no volume at all, so it used the new in-image default and worked fine. The broken artifact was the generated compose file, which nothing in CI ever stands up.

That is the real gap: verify exercises the running stack, not what init writes to disk. This PR closes the specific hole with a unit test asserting the generated mount path, but the general shape — a generated artifact no test executes — is worth a follow-up.

Verified

Scaffolded a project with the built CLI and brought its database up:

• running (healthy), PostgreSQL 18.4 (Debian 18.4-1.pgdg13+1)
• PGDATA=/var/lib/postgresql/18/docker, resolving inside the mounted volume
• wrote a row, docker compose up -d --force-recreate db, read it back: survives, so the volume is genuinely in use rather than data silently living in the container layer

Also reproduced the failure first with the old mount, so the test is against a real defect rather than a theory. npm test passes (816 passed, 4 skipped).

Release impact

None shipped. 0.11.0 is on npm; the PostgreSQL 18 change is still sitting unreleased in the release PR (#163). This lands ahead of it, so no published version ever produced a broken scaffold.

Related

The same mount change is needed everywhere else postgres moves to 18. The five issues I filed for the org rollout all say "bump the tag" and none mention the mount; I'm updating them now.

[Bccorb]

Filed #166 for the underlying gap this exposed: nothing in CI executes what init generates, so the broken scaffold passed both jobs and review. It also records a related finding, that seamless check exits 0 even when every service is unreachable, which is why it cannot be dropped in as the CI assertion without a change.