Integrate VPA in knative/serving

/area autoscale

## Describe the feature

We would love to see VPA integrated in knative, as tried before in https://github.com/knative/serving/issues/2929

### Use case

### Context

We currently have our applications (mostly web applications) deployed in kubernetes with standard k8s objects, and we are evaluating replacing such deployments with knative services.

#### Current autoscaling strategy

Our autoscaling strategy is set up as follows:
* Each deployment defines its VPA and HPA
* We use RPS as main HPA metric
* We do not use resource policy limits
* We rely on VPA to set requested usage

This implies that just by defining how many requests per second a pod is able to handle, all the rest gets done automatically.
One advantage that we find in this strategy is right-sizing the deployments almost free (even if we give conservative estimations on supported RPS, with VPA the CPU/Memory requests will be sized accordingly)
As we can appreciate, **VPA is key** to this strategy

Migrating to knative without VPA would force us to manually define resource policy for all services, thus producing either toil or over/under-provisioning with all consequences that this might bring.

At this moment, we find that knative would benefit from having VPA for all users that are in a situation similar to ours, and unfortunately this is **blocking a full rollout** of knative services to our platform.

Is there any chance that VPA integration in knative serving could be considered?