Use composite cumulator in MessageAccumulator

This seem to give great performance boost for very large tx ( > 16 MB)

We did not see any degradation for smaller tx sized.

It is possible to reset to the old cumulator through a feature toggle.

community/bolt/src/main/java/org/neo4j/bolt/transport/pipeline/MessageAccumulator.java

@@ -26,10 +26,25 @@
 
 import java.util.List;
 
+import org.neo4j.util.FeatureToggles;
+
 public class MessageAccumulator extends ByteToMessageDecoder
 {
+    private static final boolean USE_MERGE_CUMULATOR = FeatureToggles.flag( MessageAccumulator.class, "mergeCumulator", false );
     private boolean readMessageBoundary;
 
+    public MessageAccumulator()
+    {
+        if ( USE_MERGE_CUMULATOR )
+        {
+            setCumulator( MERGE_CUMULATOR );
+        }
+        else
+        {
+            setCumulator( COMPOSITE_CUMULATOR );
+        }
+    }
+
     @Override
     public void channelRead( ChannelHandlerContext ctx, Object msg ) throws Exception
     {