Avoid flooding replication source RSes logs when no sinks are available

On HBaseInterClusterReplicationEndpoint.replicate, if no sinks are returned by ReplicationSinkManager, say remote peer is not available, we log message below and return false to source shipper thread, which then keeps retrying, flooding source RS log with the below messages:

WARN org.apache.hadoop.hbase.replication.regionserver.HBaseInterClusterReplicationEndpoint: No replication sinks found, returning without replicating. The source should retry with the same set of edits.

This condition could also cause ReplicationSinkManager.chooseSinks to blow an NPE.