Implement change to voting with no leader

Implement change to voting with no leader

Backported from master.

Implemented a special case where on a voting state change from
non-voting to voting, if there's no leader, it will try to elect a
leader in order to apply the change and progress.

This is to handle a use case where one has 2 geographically-separated
3-node clusters, one a primary and the other a backup such that if the
primary cluster is lost, the backup can take over. In this scenario,
there's a logical 6-node cluster where the primary sub-cluster is
configured as voting and the backup sub-cluster as non-voting such
that the primary cluster can  make progress without consensus from
the backup cluster while still replicating to the backup. On fail-over
to the backup, a request would be sent to a member of the backup
cluster to flip the voting states, ie make the backup sub-cluster
voting and the lost primary non-voting. However since the primary
majority cluster is lost, there would be no leader to apply, persist and
replicate the server config change.

Therefore, if the server processing the request is currently non-voting
and is to be changed to voting and there is no current leader, it will
try to elect itself the leader by applying the new server config change in
the RaftActorContext and sending an ElectionTimeout. If it's elected
leader, it persists and replicates the new server config. If no leader
change occurs within the election timeout period, it reverts the server
config change and tries to forward the change request to another server
with the same voting state change. In this manner, the intent is to elect
the newly voting server that has the most up to date log.

Change-Id: I67b5b2d3a97745dbe9a8215f9a28f3a840f2a0db
Signed-off-by: Tom Pantelis <tpanteli@brocade.com>