ceph: bulk restart: reuse the health classification between steps

The three health checks inside the restart loops still used the
overall status. They would abort at the first step on any cluster that
the new entry check lets through. They also run on resume, where the
entry check is skipped.

Send them through the same classification, limited to blocking errors.
A rolling restart causes warnings by itself, PG_DEGRADED in
particular, and reporting those as the reason for an abort hides the
real cause.

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