Fix in-cluster kubectl --namespace override

Before this change, if the config was empty, ConfirmUsable() would
return an "invalid configuration" error instead of examining and
honoring the value of the --namespace flag. This change looks at the
overrides first, and returns the overridden value if it exists before
attempting to check if the config is usable. This is most applicable to
in-cluster clients, where they don't have a kubeconfig but they do have
a token and can use KUBERNETES_SERVICE_HOST/_PORT.

Kubernetes-commit: 23e32b100fb9745b70203c30716697bd03926313

tools/clientcmd/client_config_test.go

@@ -504,3 +504,25 @@ func matchByteArg(expected, got []byte, t *testing.T) {
 		t.Errorf("Expected %v, got %v", expected, got)
 	}
 }
+
+func TestNamespaceOverride(t *testing.T) {
+	config := &DirectClientConfig{
+		overrides: &ConfigOverrides{
+			Context: clientcmdapi.Context{
+				Namespace: "foo",
+			},
+		},
+	}
+
+	ns, overridden, err := config.Namespace()
+
+	if err != nil {
+		t.Errorf("Unexpected error: %v", err)
+	}
+
+	if !overridden {
+		t.Errorf("Expected overridden = true")
+	}
+
+	matchStringArg("foo", ns, t)
+}