Feature-Request: add memory utilization to activator-hpa

## Context
Currently we set activator-hpa only based on a `cpu` [target](https://github.com/knative/serving/blob/main/config/core/deployments/activator-hpa.yaml#L34). As the activator is potentially buffering a lot of requests, it can happen that the configured memory is not sufficient, causing the activator to be OOM-killed:

```
kubectl describe pod -n knative-serving activator-774d4ff4b8-4l5vp | grep Reason
Reason: OOMKilled
```

Obviously, the configured resources should be adapted depending on the expected workload (small cluster, big cluster, amount of `Kservices`, traffic payload sizes etc.), but we also could add a `memory` target to the activator-hpa to automatically scale more instances of activator if `memory` pressure increases.

## Describe the feature
- Add a `memory` target to [activator-hpa](https://github.com/knative/serving/blob/main/config/core/deployments/activator-hpa.yaml)
- Add some docs around setting resource limits based on workload sizes
- Add some docs around potential issues (like [this](https://github.com/knative/serving/issues/13583#issuecomment-1469160313)) and recommendations (like setting TBC to 0).

Related: https://github.com/knative/serving/issues/13583

/area autoscale

Start with config/core/deployments/activator-hpa.yaml and review the existing CPU target. Read the linked issue 13583 and its referenced comment for the potential issues and recommendations, then identify where workload-based resource guidance belongs. Done means the activator HPA includes a memory target and documentation covers resource limits, risks, and recommendations such as setting TBC to 0.

Written by the indexing model from the issue text.