… warnings.

An oversight when we removed non-sharding modes of operation is that the `local`
storage type stopped working. Unfortunately it is not conceptually simple to
support this type fully, as alertmanager requires remote storage shared between
all replicas, to support recovering tenant state to an arbitrary replica
following an all-replica outage.

To support provisioning of alerts with `local` storage, but persisting of state
to remote storage, we would need to allow different storage configurations.

This change fixes the issue in a more naive way, so that the alertmanager can at
least be started up for testing or development purposes, but persisting state
will always fail. A second PR will propose allowing the `Persister` to be
disabled.

Although this configuration is not recommended for production used, as long as
the number of replicas is equal to the replication factor, then tenants will
never move between replicas, and so the local snapshot behaviour of the upstream
alertmanager will be sufficient.

Fixes #1638