ceph: services: decide on per-check severity, not on cluster status

The decision was based on Ceph's overall health status. That made the
per-check allowlist useless on a HEALTH_ERR cluster: an allowlisted
check still blocked the restart, and muting only helped if every
firing check was muted.

Look at the severity of each check instead. An unknown severity counts
as HEALTH_ERR, so an unexpected value from Ceph blocks. The returned
severity now describes the checks that still block, not the cluster,
so a HEALTH_ERR cluster can have nothing that blocks a restart.

Signed-off-by: Thomas Lamprecht <t.lamprecht@proxmox.com>