LU-18973 mgc: account failovers from all sources

LU-18973 mgc: account failovers from all sources

Once set up initially MGC is not updating import failovers
from other mounts. That causes problems with MGC on MGS -
it is always set up with only @lo interface, so if MGS
failed over to other node, all targets/clients on primary
node are unable to find MGS, because MGC has only @lo peer

Patch reworks lustre_start_mgc() code to account all
failover peers from each user of that MGC. It adds new
failover NIDs even if MGC exists already.

Patch re-organizes also the way how  peers are identified.
It uses peer UUID as 'Primary NID' string instead of
naming it as 'MGC<PrimaryNID>_##' so same NIDs don't
produces new mappings and don't pollute import with
duplicated connections.

That makes LCFG_DEL_UUID obsoleted as well, because
lustre_stop_mgc() was its last user.

Signed-off-by: Mikhail Pershin <mpershin@whamcloud.com>
Change-Id: Icea5b74a16972e8a5f2737257086074630e652a8
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/59076
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Sebastien Buisson <sbuisson@ddn.com>
Reviewed-by: Marc Vef <mvef@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>