C2(deploy-infra-alerting): impl fix alert rule multi-series evaluation

Add reduce step between Prometheus query and threshold to preserve
per-service labels. Without it, Grafana can't distinguish the 5
probe_success series and errors with "duplicate results with labels {}".

Chain: A (prometheus query) → B (reduce last) → C (threshold < 1)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

argocd/manifests/grafana/alerting.yaml

@@ -34,7 +34,7 @@ groups:
     rules:
       - uid: service-probe-failure
         title: ServiceProbeFailure
-        condition: B
+        condition: C
         for: 2m
         noDataState: Alerting
         execErrState: Alerting
@@ -62,8 +62,20 @@ groups:
               from: 0
               to: 0
             model:
-              type: threshold
+              type: reduce
               expression: A
+              reducer: last
+              settings:
+                mode: dropNN
+              refId: B
+          - refId: C
+            datasourceUid: "__expr__"
+            relativeTimeRange:
+              from: 0
+              to: 0
+            model:
+              type: threshold
+              expression: B
               conditions:
                 - evaluator:
                     type: lt
@@ -71,9 +83,7 @@ groups:
                       - 1
                   operator:
                     type: and
-                  reducer:
-                    type: last
-              refId: B
+              refId: C
 
 templates:
   - orgId: 1