mgr/cephadm: batch MDS upgrades when fail_fs is set

When fail_fs is enabled the filesystem is already failed, so MDS
daemons are safe to upgrade together. Stop breaking after the first
MDS in _to_upgrade so they are queued in one pass instead of one
per upgrade cycle.

Fixes: https://tracker.ceph.com/issues/78304
Signed-off-by: Kobi Ginon <kginon@redhat.com>