Fix lost committed events by moving leader during snapshot synchronization

[negokaz]

Closes #133

Overview

• Introduce a new persisted event SnapshotSyncStarted that is written when snapshot synchronization started
  • RaftActor gets a new property lastSnapshotState.isDirty that indicate that snapshot synchronization is incomplete
  • The property is persisted state by SnapshotSyncStarted
• RaftActor doesn't accept AppendEntries if lastSnapshotState.isDirty is true
• RaftActor rejects InstallSnapshot that is sent by the leader with old snapshots than the previous leader
  • To completely overwrite a dirty snapshot changes when a snapshot synchronization fails
  • This is achieved by new properties lastSnapshotState.targetSnapshotLastTerm and lastSnapshotState.targetSnapshotLastLogIndex
  • These properties are persisted since they are updated by SnapshotSyncStarted
• Starting snapshot synchronization resets RaftActor's log
  • This allows quick recovery for the delayed follower by adjusting the leader's nextIndex during the synchronization
• This PR changes the candidate behavior as well as follower behavior since the follower may become a candidate by ElectionTimeout
  • We don't have to change leader behavior since the delayed follower can not be a leader
• This change may reduce the availability instead of data consistency
  • Members that are synchronizing snapshots can not contribute to committing events