ceph: services: only clear the noout flags a bulk restart set itself

with_noout() set the flag on every OSD it was given and cleared it on
every one again, without checking which OSDs already had it. An
operator who had set noout on a single OSD, to stop it from flapping
or while working on its disk, silently lost that flag as soon as any
rolling restart finished.

Read the per-OSD flag from 'osd dump' first and only touch the OSDs
that do not have it yet. This makes the flag loss less likely, but it
does not remove it: a flag that somebody sets while the restart runs
is still cleared at the end, and the OSD map does not record who set a
flag.

The ids whose flag this run set are handed to the caller right away,
because the cleanup here cannot survive a SIGKILL. The bulk restart
stores them with its progress and passes them back on resume, where a
fresh ownership check would find them already set and claim none of
them.

The two callers used different id formats, 'osd.N' against a bare N,
so normalise before comparing against the OSD map.

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