Fix Debian PostgreSQL setup: create cluster if missing and target version-specific systemd unit

[nuz014]

Summary

Two related fixes for running this role against Debian/Ubuntu hosts, surfaced while
installing PostgreSQL 18 (incl. side-by-side with an existing cluster).

1. Ensure the Debian/Ubuntu cluster exists before configuring it

The role assumed the postgresql-NN package's postinst had already created the
main cluster via pg_createcluster. When that auto-creation is skipped or fails
(locale issues, create_main_cluster=false in createcluster.conf, or the config
dir already existing), /etc/postgresql/NN/main/postgresql.conf is never written and
the role fails with a confusing error two tasks later:

TASK [galaxyproject.postgresql : Check for conf.d include in postgresql.conf]
Destination /etc/postgresql/18/main/postgresql.conf does not exist !

Worse, the Create conf.d task silently creates the empty /etc/postgresql/NN/main/
parent directory, which then blocks a later pg_createcluster and makes the failure
sticky across re-runs.

This PR adds an explicit, idempotent cluster-creation step to tasks/debian.yml
(guarded by creates:) that runs before Create conf.d, so the cluster is always
in place before the role writes any config into it. On hosts where the cluster already
exists it is a no-op.

2. Target the version-specific systemd unit on Debian instead of the meta-service

On Debian, postgresql_service_name was the postgresql meta-service. That oneshot is
already active whenever any cluster is running (e.g. an existing 15/main), so:

• service: name=postgresql state=started is a no-op and never starts the cluster
  this role is managing — the new cluster stays down after the run.
• the Reload PostgreSQL handler (state=reloaded) reloads every cluster on the
  host, unexpectedly touching unrelated/production clusters.

This PR changes the Debian default to postgresql@{{ postgresql_version }}-main,
mirroring what RedHat already does (postgresql-{{ postgresql_version }}). Start/enable
and reload now affect only the cluster this role manages, which also makes side-by-side
installs (e.g. 15 and 18 on one host) behave correctly.

Changes

• tasks/debian.yml: add Ensure the PostgreSQL cluster exists (Debian) using
  pg_createcluster {{ postgresql_version }} main with a creates: guard.
• vars/debian.yml: postgresql_service_name → postgresql@{{ postgresql_version }}-main.

Testing

• Fresh Debian/Ubuntu host: cluster is created, configured, and started by the role.
• Host with an existing cluster (e.g. 15/main running): installing a second version
  (18) now creates, configures, starts the 18 cluster on its own port, and leaves
  the existing cluster untouched (no cross-cluster reload).
• Re-runs are idempotent (cluster creation is a no-op once postgresql.conf exists).

Notes / compatibility

• No change for RedHat (already version-specific).
• Single-cluster Debian hosts are unaffected in behavior; targeting the cluster unit is
  equivalent to the meta-service when only one cluster exists.

[nuz014]

resolves #23 and also #7