Stabilize RareClusterState

* Use actual master node, not just a master elligible node when trying to cancel publication. This only works on the master and for unlucky seeds we never try the master within the 10s that the busy assert runs.
* Closes #36813

server/src/test/java/org/elasticsearch/cluster/coordination/RareClusterStateIT.java

@@ -146,7 +146,8 @@ public void onFailure(String source, Exception e) {
     private <Req extends ActionRequest, Res extends ActionResponse> ActionFuture<Res> executeAndCancelCommittedPublication(
             ActionRequestBuilder<Req, Res> req) throws Exception {
         ActionFuture<Res> future = req.execute();
-        assertBusy(() -> assertTrue(((Coordinator)internalCluster().getMasterNodeInstance(Discovery.class)).cancelCommittedPublication()));
+        assertBusy(
+            () -> assertTrue(((Coordinator)internalCluster().getCurrentMasterNodeInstance(Discovery.class)).cancelCommittedPublication()));
         return future;
     }
 
@@ -276,7 +277,6 @@ public void testDelayedMappingPropagationOnPrimary() throws Exception {
         });
     }
 
-    @AwaitsFix(bugUrl="https://github.com/elastic/elasticsearch/issues/36813")
     public void testDelayedMappingPropagationOnReplica() throws Exception {
         // This is essentially the same thing as testDelayedMappingPropagationOnPrimary
         // but for replicas