When a test wants to create a cache manager in another thread (e.g. with fork()) it should be able to call TestResourceTracker.testThreadStarted(this) and join the same cluster.

That works with TCP, but with UDP each thread gets its own multicast address and port, so multicast messages from one node don't reach the other. Unicast messages and TEST_PING do work.

This is the issue that made FunctionalTxTest fail in https://github.com/infinispan/infinispan/pull/6442