Conflict resolution should be distributed amongst primary replicas

Currently conflict resolution is executed entirely on the cluster coordinator on a per segment basis. However, as conflict resolution requires the retrieval of all copies of a given segment to be returned, it's only possible to process entries one segment at a time in an attempt to avoid the coordinator running out of memory. Distributing CR to each segment's primary owner will allow for greater parallelism and scalability as the size of the cluster increases.