Dead peer removal logic and restart handling about constellation

[buckysballs]

The cluster needs to automatically remove peers when they become unresponsive or misaligned so we can prevent forks and free up space for new peers to join. In the case that the cluster forks (misalignment amongst the top ranked peers), we need to restart the cluster and roll back to the most recent non-forked snapshot. In the case of an unresponsive or misaligned ordinary peer, the peer should be removed from the cluster.

In the case of a rollback, we should restart the cluster with the highest reputable peers (we will guarantee a fixed set of nodes at the highest reputation) and re-add the rest of the peers that are responsive to the restart flow.

Goal

1. A restart mechanism for forks amongst highest ranked peers (tip height misalignment).
2. A removal mechanism for peers that are misaligned or unresponsive.

Definition of Done:

1. Artificially induced misalignment amongst the highest ranked nodes causes a restart/rollback the the most recent aligned snapshot
2. Artificially induced misalignment amongst ordinary peers causes the peer to be removed from the cluster.

[wkoszycki]

connected with #603