Move logging shutdown to after waiting on node close (elastic#95093)

When Elasticsearch is shutting down, the Node instance is closed, and
then logging is shutdown. However, some aspects of Node shutdown happen
asynchronously, and awaiting that final close from the node occurs after
shutting down logging. This commit moves logging shutdown to the very
end of shutdown, to allow for as much information as possible to be
logged if something goes wrong.

Author: rjernst

server/src/main/java/org/elasticsearch/bootstrap/Elasticsearch.java

@@ -446,8 +446,6 @@ private static void shutdown() {
         var es = INSTANCE;
         try {
             IOUtils.close(es.node, es.spawner);
-            LoggerContext context = (LoggerContext) LogManager.getContext(false);
-            Configurator.shutdown(context);
             if (es.node != null && es.node.awaitClose(10, TimeUnit.SECONDS) == false) {
                 throw new IllegalStateException(
                     "Node didn't stop within 10 seconds. " + "Any outstanding requests or tasks might get killed."
@@ -459,6 +457,9 @@ private static void shutdown() {
             LogManager.getLogger(Elasticsearch.class).warn("Thread got interrupted while waiting for the node to shutdown.");
             Thread.currentThread().interrupt();
         } finally {
+            LoggerContext context = (LoggerContext) LogManager.getContext(false);
+            Configurator.shutdown(context);
+
             es.keepAliveLatch.countDown();
         }
     }