[Test] Flush master queue before checking snapshots (elastic#116938) (elastic#122721)

The block-on-data-node returns once the data node begins to process the
cluster state update for new snapshot. This is before master can see the
chnages. In edge cases, the listener may be completed too early before
the master can see the new snapshot. This PR flushes the master queue to
ensure the snapshot is visible.

Resolves: elastic#116730
(cherry picked from commit 5d9385f)

# Conflicts:
#	muted-tests.yml

Author: ywangd

server/src/internalClusterTest/java/org/elasticsearch/snapshots/SnapshotShutdownIT.java

@@ -72,6 +72,7 @@ public void testRestartNodeDuringSnapshot() throws Exception {
 
         final var clusterService = internalCluster().getCurrentMasterNodeInstance(ClusterService.class);
         final var snapshotFuture = startFullSnapshotBlockedOnDataNode(randomIdentifier(), repoName, originalNode);
+        safeAwait((ActionListener<Void> l) -> flushMasterQueue(clusterService, l));
         final var snapshotCompletesWithoutPausingListener = ClusterServiceUtils.addTemporaryStateListener(clusterService, state -> {
             final var entriesForRepo = SnapshotsInProgress.get(state).forRepo(repoName);
             if (entriesForRepo.isEmpty()) {