Scrape_configs: retain cAdvisor metrics for system processes

Currently, CPU metrics for Kubelet, Containerd, etc. are dropped by a
relabel rule since they don't have a container image.

On most systems they run in their own cgroup called `system.slice`, so
we can add another rule which uses that fact to give the `image` label a
value and then they won't be dropped.

Author: bboreham

prometheus/scrape_configs.libsonnet

@@ -255,6 +255,14 @@
     ],
 
     metric_relabel_configs: [
+      // Let system processes like kubelet survive the next rule by giving them a fake image.
+      {
+        source_labels: ['__name__', 'id'],
+        regex: 'container_([a-z_]+);/system.slice/(.+)',
+        target_label: 'image',
+        replacement: '$2',
+      },
+
       // Drop container_* metrics with no image.
       {
         source_labels: ['__name__', 'image'],