Alertmanager cluster CLI flags have the wrong prefix

[pracucci]

Describe the bug
The multitenant alertmanager CLI flags used to configure the alertmanager cluster are missing the -alertmanager. prefix (they start with -cluster. which is quite generic and may lead to misunderstandings):

cortex/pkg/alertmanager/multitenant.go

Lines 118 to 121 in e4867d8

	f.StringVar(&cfg.ClusterBindAddr, "cluster.listen-address", defaultClusterAddr, "Listen address for cluster.")
	f.StringVar(&cfg.ClusterAdvertiseAddr, "cluster.advertise-address", "", "Explicit address to advertise in cluster.")
	f.Var(&cfg.Peers, "cluster.peer", "Initial peers (may be repeated).")
	f.DurationVar(&cfg.PeerTimeout, "cluster.peer-timeout", time.Second*15, "Time to wait between peers to send notifications.")

According to the v1 guarantees, only the alertmanager APIs are marked experimental, so we can't change these CLI flags straightaway but we need to introduce the new flags, deprecate the old ones (while keeping backward compatibility) and remove the old ones after 2 releases.

[gotjosh]

While we're at this, we need to change cluster.peer to accept multiple peers as part of its input. The model where we need to repeat the flag several types does not work well with our integration test framework that assumes flags are a map.

We have flagext.StringSliceCSV for these purposes.