[raft] Improve startup and joining procedure #1802

Following feedback on #1798 and #1780, design a startup and join procedure that is k8s friendly and avoids paths where trivial mistakes lead to a silently failing cluster.

Design proposal according to the feedback

• Every node has a unique name within the cluster (TBD - use locality / public endpoint / other ?)
• The name is replicated so that the node can still be uniquely identifiable within the cluster
• It has 3 Raft IDs (one per service)
• The IDs are autogenerated at the start and saved in the WAL's metadata

Starting up and handling membership

Bootstrapping a cluster

• Every node starts with its own URL only
• Nodes don't find a wal, they generate an ID per service
• They wait
• User runs dssctl bootstrap --node A (new admin endpoint)
• Node A runs raft.StartNode and creates a new cluster with itself as a leader
• Add the other nodes one by one to the cluster with dssctl add

Restarting a node

• The node finds its old wal and restarts normally

New node joining and existing cluster

• Start node then add it with dssctl add like the last step of the bootstrap procedure

Resetting a node that has lost or corrupted its datadir

• Delete the datadir
• Node starts, generates a new ID and waits
• Old ID can be removed and the new one added

(This design removes all state-dependent flags and ensures that, either nodes know what to do at startup or that they will wait for a signal on the admin API from dssctl)