ceph: services: let the bulk restart lock cover a long multi-type run

A caller that goes through more than one daemon type has to block
other per-type restarts for the whole run. The lock now accepts
several scopes at once instead of one. If one of them cannot be taken,
the ones already held are released again.

The lock entry also goes stale after four hours, which going through
every OSD of a large cluster can exceed, and another bulk restart then
treats it as abandoned. Accept the upid of the current holder as a
renewal, so a long run can refresh its entry while it works. A crashed
run cannot renew, as every run builds a new upid, so its entry is only
freed once it goes stale.

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