Use a separate ClusterParticipant for Stats Aggregation in Ambry

[jsjtzyy]

Currently the stats aggregation relies on HelixParticipant and sort of depends on same ZKHelixManager. A separate ClusterParticipant and an independent ZKHelixManager could be set up for stats aggregation. Thus, stats aggregation could work no matter StaticClusterManager or HelixClusterManager is being used. Another benefit we may get is, for HelixParticipant, it can participate earlier which may avoid race conditions. (Currently we add extra lock to eliminate such race condition)