[abd-ulbasit]

Closes #966

The PrometheusRule template builds podSelector as
<fullname>-([1-9][0-9]*)$. The low disk space rules append -wal and
-tbs.* to it, so they render as:

persistentvolumeclaim=~"<fullname>-([1-9][0-9]*)$-wal"
persistentvolumeclaim=~"<fullname>-([1-9][0-9]*)$-tbs.*"

$ only matches at the end of the value, so neither matcher can match
anything. A full WAL or tablespace volume raises neither
CNPGClusterLowDiskSpaceWarning nor CNPGClusterLowDiskSpaceCritical,
although the runbooks say both volumes are covered. This goes back to
the first version of the rules (#193).

The fix drops the $. Prometheus anchors =~ matchers at both ends
anyway, so the other matchers built from podSelector (the pod=~
matchers and the PGDATA persistentvolumeclaim=~ matchers) select the
same series as before. In the rendered monitoring test, these matchers
are the only lines that change.

The monitoring chainsaw test now checks that both low disk space rules
render the -wal and -tbs.* matchers without the $.

Testing:

• I did not run the full chainsaw suite (no cluster on my side). I ran
  the new assertions with chainsaw assert -r (v0.2.15) against the
  helm template output. They pass with the fix. All four fail on main.
• helm lint charts/cluster
• ct lint --target-branch=main --check-version-increment=false (ct v3.14.0)
• promtool test rules (v3.15.1) on the rendered rules. With a full WAL
  PVC or a full tablespace PVC, both alerts fire with the fix and stay
  silent on main. With no full PVC, nothing fires. Without the $,
  CNPGClusterInstancesOnSameNode still ignores pooler and job pods on
  the same node.

Disclosure: parts of this change and this description were written with
Claude Code (Claude Opus 5.5), as noted in the Assisted-by commit
trailer.