TransportService should honor handler executor on failure

[dnhatn]

There are instances where the TransportService may invoke onFailure using a thread other than the one specified in the handler executor.

at org.elasticsearch.action.ActionListenerImplementations.safeAcceptException(ActionListenerImplementations.java:62)
at org.elasticsearch.action.ActionListener$1.onFailure(ActionListener.java:140)
at org.elasticsearch.action.ActionListenerResponseHandler.handleException(ActionListenerResponseHandler.java:53)
at org.elasticsearch.transport.TransportService.handleSendRequestException(TransportService.java:859)
at org.elasticsearch.transport.TransportService.getConnectionOrFail(TransportService.java:781)
at org.elasticsearch.transport.TransportService.sendRequest(TransportService.java:757)
at org.elasticsearch.transport.TransportService.sendRequest(TransportService.java:747)

This also happens with RemoteAwareClient:

at org.elasticsearch.action.ActionListenerImplementations.safeAcceptException(ActionListenerImplementations.java:62)
at org.elasticsearch.action.ActionListenerImplementations.safeOnFailure(ActionListenerImplementations.java:73)
at org.elasticsearch.action.DelegatingActionListener.onFailure(DelegatingActionListener.java:31)
at org.elasticsearch.action.ActionListenerImplementations$ResponseWrappingActionListener.onResponse(ActionListenerImplementations.java:247)
at org.elasticsearch.transport.RemoteClusterAwareClient.maybeEnsureConnected(RemoteClusterAwareClient.java:71)
at org.elasticsearch.transport.RemoteClusterAwareClient.execute(RemoteClusterAwareClient.java:41)

[elasticsearchmachine]

Pinging @elastic/es-distributed (Team:Distributed)