Deduplicate generated alerts in HA setup

Removes the instance-specific 'monitor' label set on alerts, so that
multiple Prometheus instances in a HA setup all create the same
alerts.

roles/prometheus/templates/alertmanager.yml.j2

@@ -13,7 +13,7 @@ route:
   group_by: ['alertname', 'service']
   group_wait: 30s
   group_interval: 5m
-  repeat_interval: 3h
+  repeat_interval: 12h
   receiver: default
 
 # The following inhibit rules are meant to work both with the job/host

roles/prometheus/templates/prometheus.yml.j2

@@ -64,10 +64,17 @@
 
 global:
   scrape_interval: "10s"
+
+  # Set an external label unique to this host, but remove it
+  # from the generated alerts so that they match across all
+  # Prometheus instances.
   external_labels:
     monitor: "{{ inventory_hostname }}"
 
 alerting:
+  alert_relabel_configs:
+    - regex: monitor
+      action: labeldrop
   alertmanagers:
     - static_configs:
 {{ targets_for_service('prometheus', 'prometheus', 9093) }}