Optimal config for low latency

The default bundler TransferQueueBundler is biased towards high throughput, but not necessarily for low latency.
Investigate a configuration and/or changes to provide low latency, e.g. for synchronous RPCs. This includes:

• Changes to TransferQueueBundler (see TransferQueueBundler2)
• Use of BATCH, especially for smaller messages
• New flag LOW_LATENCY, which can be handled differently by protocols. For example, flow control might bypass messages with this flag
• Removal of UFC / MFC for low latency messages (see above)
• Increasing ack_threshold im UNICAST3
• Use of async invocation handlers: http://www.jgroups.org/manual5/index.html#AsyncInvocation
• Document: https://github.com/belaban/JGroups/blob/master/doc/design/LatencyVersusThroughput.txt