_Extracts head or tail outside compilation to separate host launches before/after device cluster._

This pass extracts a CPU computation cluster with `_xla_outside_compilation`
annotation from the head or tail of a Device cluster.

For example:

```mlir
  %cluster = "tf_device.cluster"() ( {
    %a = "tf.A"(%arg0) {_xla_outside_compilation = "cluster1"} : (tensor<i32>) -> tensor<i32>
    %b = "tf.B"(%a) : (tensor<i32>) -> tensor<i32>

func (c Clusters) Default() Cluster {
	return c[0]
}

// GetOrDefault returns the given cluster if non-nil. Otherwise returns the first
// Cluster in the list.
func (c Clusters) GetOrDefault(cluster Cluster) Cluster {
	if cluster != nil {
		return cluster
	}
	return c.Default()
}

// GetByName returns the Cluster with the given name or nil if it is not in the list.

					model.EndpointType: {
						NonceSent:     "nonce",
						ResourceNames: []string{"cluster2", "cluster1"},
					},
				},
			},
			request: &discovery.DiscoveryRequest{
				TypeUrl:       model.EndpointType,
				VersionInfo:   "v1",
				ResponseNonce: "nonce",
				ResourceNames: []string{"cluster1", "cluster2"},
			},
			response: false,
		},
		{
			name: "unsubscribe EDS",

running multiple k8s clusters it is important to proivde different values for each cluster
to avoid overlapping addresses.

For example, to create two clusters, run the script two times with the following
parameters:

```bash
  ./setupkind.sh --cluster-name cluster1 --ip-space 255 -i dual
  ./setupkind.sh --cluster-name cluster2 --ip-space 245 -i dual
```

		{Name: "inbound|7443||", DnsLookupFamily: cluster.Cluster_V4_ONLY, LbPolicy: cluster.Cluster_ROUND_ROBIN},
	}
	sidecarInboundServiceOut := []*cluster.Cluster{
		{
			Name: "inbound|7443||", ClusterDiscoveryType: &cluster.Cluster_Type{Type: cluster.Cluster_EDS},
			DnsLookupFamily: cluster.Cluster_V6_ONLY, LbPolicy: cluster.Cluster_RING_HASH,
		},
	}

	gatewayInput := []*cluster.Cluster{

			input: map[string]*discovery.DiscoveryResponse{
				"istiod1": xdsResponseInput("istiod1", []clientConfigInput{
					{
						proxyID:        "proxy1",
						clusterID:      "cluster1",
						version:        "1.20",
						cdsSyncStatus:  status.ConfigStatus_STALE,
						ldsSyncStatus:  status.ConfigStatus_SYNCED,
						rdsSyncStatus:  status.ConfigStatus_NOT_SENT,

		{"cluster-1", cluster2, true},
		// always search external
		{"cluster-1", external, false},
		{"cluster-2", external, false},
		{"", external, false},
		// always search for empty node cluster id
		{"", cluster1, false},
		{"", cluster2, false},
		{"", external, false},
	}

	for i, c := range cases {
		got := skipSearchingRegistryForProxy(c.nodeClusterID, c.registry)
		if got != c.want {

	"istio.io/istio/pilot/pkg/features"
	"istio.io/istio/pkg/cluster"
	"istio.io/istio/pkg/config/mesh"
	"istio.io/istio/pkg/kube"
	"istio.io/istio/pkg/kube/kclient"
	filter "istio.io/istio/pkg/kube/namespace"
	"istio.io/istio/pkg/log"
)

// Cluster defines cluster struct
type Cluster struct {
	// ID of the cluster.
	ID cluster.ID
	// Client for accessing the cluster.
	Client kube.Client

	kubeConfigSha [sha256.Size]byte

	}
	diff := difflib.UnifiedDiff{
		FromFile: "Istiod Clusters",
		A:        difflib.SplitLines(istiodBytes.String()),
		ToFile:   "Envoy Clusters",
		B:        difflib.SplitLines(envoyBytes.String()),
		Context:  c.context,
	}
	text, err := difflib.GetUnifiedDiffString(diff)
	if err != nil {
		return err
	}
	if text != "" {
		fmt.Fprintln(c.w, "Clusters Don't Match")
		fmt.Fprintln(c.w, text)
	} else {

	cases := []struct {
		name    string
		cluster cluster.ID
		network network.ID
	}{
		{
			name:    "values unset",
			cluster: "",
			network: "",
		},
		{
			name:    "values set",
			cluster: testC,
			network: testNW,
		},
	}

	for _, c := range cases {
		t.Run(c.name, func(t *testing.T) {
			s := newAmbientTestServer(t, c.cluster, c.network)