[fix](cloud) Fixed uneven tablet performance during upgrades from older versions (#58135)

Prior to this #42986 PR, the
`getLastUpdateMs` field of the backend was not written to the fe image.
This caused the tablet mapping on the cloud to lose the PrimaryBackend
information after the upgrade, resulting in the regeneration of an
incorrect PrimaryBackend and uneven tablet distribution.

detail:
<img width="930" height="859" alt="image"
src="https://github.com/user-attachments/assets/23d62adc-0bbb-49cb-b5b0-d1768f6422c7"
/>

Author: deardeng

fe/fe-core/src/main/java/org/apache/doris/cloud/catalog/CloudTabletRebalancer.java

@@ -651,7 +651,9 @@ private boolean completeRouteInfo() {
                     // primary backend is alive or dead not long
                     Backend be = replica.getPrimaryBackend(cluster, false);
                     if (be != null && (be.isQueryAvailable()
-                            || (!be.isQueryDisabled() && be.getLastUpdateMs() > needRehashDeadTime))) {
+                            || (!be.isQueryDisabled()
+                            // Compatible with older version upgrades, see https://github.com/apache/doris/pull/42986
+                            && (be.getLastUpdateMs() <= 0 || be.getLastUpdateMs() > needRehashDeadTime)))) {
                         beIds.add(be.getId());
                         tabletIds.add(tablet.getId());
                         continue;