(Addons) Create a new command in every retry

Whenever a `kubectl apply` fails while enabling an addon, it is
retried with exponential backoff. The command (type `*exec.Cmd`)
that this retry function runs in created outside the function -
which means that it is reused on every retry. This is a problem
because `exec.Cmd` (https://godoc.org/github.com/pkg/exec#Cmd)
states that "... Cmd cannot be reused after calling its Run or
Start methods."

This retry is a common case due to, say, a CRD and its resource
being present in the same YAML file of the addon which causes
a race condition where the resource is created before its CRD is
created in the cluster - this race is fixed by subsequent retries.
I've noticed this in the dashboard and the ambassador addon.

Due to the above mentioned bug, minikube throws errors like
`exec: already started` in every retry and the retry is never
successful, manifests are never deployed and addon creation errors
out.

Fix kubernetes#8138
Fix kubernetes#8119
Fix a few CI errors in kubernetes#8372

pkg/addons/addons.go

@@ -250,11 +250,9 @@ func enableOrDisableAddonInternal(cc *config.ClusterConfig, addon *assets.Addon,
 		}
 	}
 
-	command := kubectlCommand(cc, deployFiles, enable)
-
 	// Retry, because sometimes we race against an apiserver restart
 	apply := func() error {
-		_, err := cmd.RunCmd(command)
+		_, err := cmd.RunCmd(kubectlCommand(cc, deployFiles, enable))
 		if err != nil {
 			glog.Warningf("apply failed, will retry: %v", err)
 		}