VPA: Add benchmarking for recommender + admission-controller

[maxcao13]

Which component are you using?:

/area vertical-pod-autoscaler

Is your feature request designed to solve a problem? If so describe the problem this feature should solve.:

The current benchmarking code in vertical-pod-autoscaler/benchmark currently only displays step latencies benchmarks for the updater component. We should extend this to include the recommender and the admission-controller.

Describe the solution you'd like.:

Similar benchmarking steps and output tables for the recommender + admission-controller components as the updater.

We should probably wait and scrape recommender metrics during this section of the code:

autoscaler/vertical-pod-autoscaler/benchmark/main.go

Lines 300 to 318 in 2ceb540

	// Step 7: Scale up recommender (not updater yet)
	fmt.Println("Scaling up recommender...")
	if err := scaleUpRecommender(ctx, kubeClient); err != nil {
	return nil, fmt.Errorf("failed to scale up recommender: %v", err)
	}
	// Step 8: Wait for VPA recommendations
	fmt.Println("Waiting for VPA recommendations...")
	wait.PollUntilContextTimeout(ctx, 5*time.Second, 2*time.Minute, true, func(ctx context.Context) (bool, error) {
	vpas, _ := vpaClient.AutoscalingV1().VerticalPodAutoscalers(benchmarkNamespace).List(ctx, metav1.ListOptions{})
	withRec := 0
	for _, v := range vpas.Items {
	if v.Status.Recommendation != nil {
	withRec++
	}
	}
	fmt.Printf(" VPAs with recommendations: %d/%d\n", withRec, count)
	return withRec == count, nil
	})

(while we are waiting for recommendations).

For the admission-controller, it might make sense for it to be running from beginning to end, and we should calculate it's per-request average time.

Describe any alternative solutions you've considered.:

Additional context.:

[maxcao13]

/assign

[omerap12]

/triage accepted