Fix bug introduced in rev 1.556: the protected node thead counter

(node_threads_running) wasn't being re-initialized upon recycle, so the node
would lock/crash attempting to lock an uninitialized mutex upon answering an
incoming connection.