Hang when modifying the cluster state and an uncaught exception is thrown

[javanna]

As mentioned in #3363 we don't decrement the CountDownLatch and never return a response if there's an uncaught exception in some of the TransportMasterNodeOperationActions that modify the cluster state. Some actions do catch Throwable and don't suffer from this issue, while some others don't.

We need to go over those and fix this issue so that we always decrement the CountDownLatch and give back the error that was thrown instead of hanging waiting for the thread to finish its execution.

[javanna]

Solved with 4930b93 .

Removed CountDownLatch in all the master operations and introduced a global Listener that is always notified when something goes wrong.