Bug 2187: Bootstrap EOS shard when no local shards configured

The intended workflow to initially form a cluster dynamically is to
change the role for a second node to say member-2. Since the initial
static shard config is bootstrapped to member-1, no local shards will be
created. However, the entity-ownership shard is special in that it is
intended to exist on every node.

The EOS will be boostrapped as follows:

For the EOS CreateShard message, all unique members for all shards are
obtained from the static shard config. It assumes the local member is
present in the config however in the above workflow it won't be. So on EOS
CreateShard, if the local member isn’t in the initial member list then
it will create the local shard with an empty peer list and
DisableElectionsRaftPolicy so it stays as follower. Also the shard will
be flagged as inactive in the ShardManager. A subsequent
AddShardReplica will be needed to make it active.

The other option is to not create EOS shard but there may be initial
candidate registrations which would be missed unless we add retry logic
in the service class. But the EOS shard already has retry logic so it
would be ideal to leverage it.

I also made changes to the AddShardReplica logic to handle an existing
local shard as will occur for the EOS shard:

 - remove the failure reply if local shard already exists
 - if the local shard exists and the primary shard is the local shard,
   do nothing and return AlreadyExistsException failure reply
 - otherwise send AddServer to the primary
 - on FindPrimary, if the local shard exists but is not active, do a
   remote find as if the local shard doesn't exist
 - on AddServer, if the new server is already in the peer list, the
   ALREADY_EXISTS status is returned. Return AlreadyExistsException
   failure reply
 - on AddServer failure, if the local shard was pre-existing don't
   remove it.

We still want to prevent an AddShardReplica request which one is already
in progress so I added a Set to track this.

I added an integration test for bootstrapping the EOS shard. It starts
with an inactive shard and registers a candidate, which gets queued
since there's no leader. It then issues AddShardReplica and verifies the
candidate gets registered with the leader.

To get this to work required some teaks in the RaftActor and Follower.
When the ShardManager clears the DisableElectionsRaftPolicy, the
RaftActor creates a new Follower instance however it loses the previous
leader Id. If the new server config hasn't been replicated yet then it
has no peers and immediately tries to start an election. Since it has no
peers it goes to Leader wih no followers creating a 2 leader situation.
To alleviate this I transferred the previous leader Id to the new
Follower instance to prevent the immediate election.

Eeven after that the test still didn't work b/c the leader was still not
in the EOS shard peer list so lookup of the leader address returned
null. So I changed getPeerAddress in the RaftActorContext to lookup in
the resolver if no peer info exists.

I also added more units for AddShardReplica to increase code coverage.

Change-Id: Id2a12ae226af69611d5ca5155f5f018cef82dff4
Signed-off-by: Tom Pantelis <tpanteli@brocade.com>