Revert incorrect guard change from the previous NetworkRateBackfill fix

nics.network_rate is an unsigned int column, so it cannot store -1;
NULL is the established convention for "unlimited" there, exactly
matching how the live precedence code (NetworkOrchestrator.allocateNic())
already stores it. Changing the backfill's guard to write -1 caused a
real "Data truncation: Out of range value" failure, confirmed via a
second real upgrade run. The original rate > 0 check (skip writing on
unlimited, leaving NULL) was correct; only the router public-NIC NPE
fix from the previous commit was a genuine bug.

Author: sudo87

engine/schema/src/main/java/com/cloud/upgrade/NetworkRateBackfill.java

@@ -90,7 +90,7 @@ private void backfillNicNetworkRates() {
                 final boolean defaultNic = rs.getBoolean("default_nic");
                 try {
                     final Integer rate = computeLegacyNicNetworkRate(networkId, instanceId, defaultNic);
-                    if (rate != null && rate != 0) {
+                    if (rate != null && rate > 0) {
                         updateNicNetworkRate(nicId, rate);
                     }
                 } catch (Exception e) {