Are the Autoscaler Concurrency Metrics per pod?

## Ask your question here:
Hello, going over the [Metics documentation](https://knative.dev/docs/serving/observability/metrics/serving-metrics/#autoscaler), I have a question on the `stable_request_concurrency` and `panic_request_concurrency` metrics. The definition says "Average of requests count per observed pod ...". However, it appears that these metrics represent the total request concurrency for the revision and are not exactly the value per pod.

I tried to explore the code and this is what I found:
* [Summing the metric](https://github.com/knative/serving/blob/knative-v1.3.2/pkg/autoscaler/metrics/stats_scraper.go#L353)
* [Average](https://github.com/knative/serving/blob/knative-v1.3.2/pkg/autoscaler/metrics/collector.go#L411) - so the average is extrapolated from the sample number of pods to the total number of pods

* So, if `target_concurrency_per_pod` is `1`, there were 2 pods, and the `stable_request_concurrency` is `1.5`, we are still below the target. Is this understanding correct?
* If the above is correct, can the documentation be updated from "Average of requests count per observed pod ..." to "Estimated requests count over all observed pods ..." (or something along those lines), to avoid confusion?

Thanks!

Start with the autoscaler metrics documentation and inspect the linked code in pkg/autoscaler/metrics/stats_scraper.go and pkg/autoscaler/metrics/collector.go. Confirm whether stable_request_concurrency and panic_request_concurrency are estimated across observed pods, then update the metric definition to remove the ambiguity and clarify the target_concurrency_per_pod example.

Written by the indexing model from the issue text.