Introduce resource.PartialObject and k8s.DeferredNegotiatedSerializer…

[IfSentient]

Introduce resource.PartialObject and k8s.DeferredNegotiatedSerializer to allow for deferring more expensive unmarshaling for a given kind. This is an experimental feature to see if this accelerates initial start-up processing of large numbers of objects in CPU-constrained environments, and may be removed in a future release if not effective.

Includes changes from not-yet-merged #1023 to allow specifying the negotiated serializer using k8s.ClientConfig.

[radiohead]

A couple of minor comments, but also, could we add a benchmark? Something like:

// resource/partialobject_test.go
func BenchmarkPartialObjectUnmarshal(b *testing.B) {
    data := []byte(`{"apiVersion":"v1","kind":"Pod","metadata":{"name":"test","namespace":"default"},"spec":{"containers":[...]}}`)
    for i := 0; i < b.N; i++ {
        var obj PartialObject
        _ = json.Unmarshal(data, &obj)
    }
}

func BenchmarkTypedObjectUnmarshal(b *testing.B) {
    data := []byte(`{"apiVersion":"v1","kind":"Pod","metadata":{"name":"test","namespace":"default"},"spec":{"containers":[...]}}`)
    for i := 0; i < b.N; i++ {
        var obj TypedSpecObject[PodSpec]
        _ = json.Unmarshal(data, &obj)
    }
}

[radiohead]

Suggested change

	k8s.NewClientRegistry(config.KubeConfig, k8s.DefaultClientConfig())
	clients = k8s.NewClientRegistry(config.KubeConfig, k8s.DefaultClientConfig())

[radiohead]

Note, that this should go away after syncing with main, since #1023 was merged.

[radiohead]

These could be nil, couldn't they?

[radiohead]

@IfSentient this didn't get any traction, should we drop it?