fix: add null cluster guard before state check in CephKvmExtension

P1: Separate null check from state check. When findByUuid returns null
(cluster deleted during race with reconnect callback), the original
`cluster != null && cluster.getState() != Enabled` short-circuits false
and falls through to createSecret(), causing downstream errors. Now null
cluster triggers early return with warn-level log.

P2: Improve comment to describe the constraint technically instead of
only referencing the ticket key.

Resolves: ZSTAC-80275

Author: zstack

plugin/ceph/src/main/java/org/zstack/storage/ceph/primary/CephKvmExtension.java

@@ -58,9 +58,14 @@ public void connectionReestablished(HostInventory inv) throws HostException {
             return;
         }
 
-        // skip Ceph storage operations if cluster is not Enabled (ZSTAC-80275)
+        // skip Ceph secret creation when cluster is missing or not Enabled
         ClusterVO cluster = dbf.findByUuid(inv.getClusterUuid(), ClusterVO.class);
-        if (cluster != null && cluster.getState() != ClusterState.Enabled) {
+        if (cluster == null) {
+            logger.warn(String.format("skip Ceph secret creation on host[uuid:%s] because cluster[uuid:%s] not found",
+                    inv.getUuid(), inv.getClusterUuid()));
+            return;
+        }
+        if (cluster.getState() != ClusterState.Enabled) {
             logger.info(String.format("skip Ceph secret creation on host[uuid:%s] because cluster[uuid:%s] is %s",
                     inv.getUuid(), inv.getClusterUuid(), cluster.getState()));
             return;