Make queue-proxy NOT count the health check requests by the users

## Describe the feature
We leverage a platform managing all the knative services. All knative services report the health status via a separate URL/path, and the path is NOT consistent across different knative services.(Just confirmed internally with the team)

For example, the health check path of the knative service is determined by the model deployed in the knative service. Different models have different url for health check.

Each time, when the health check is called, the request is counted by the queue-proxy, which makes the metrics data NOT accurate, since all the health check requests are counted as well.

We are thinking of whether knative serving can distinguish the normal requests and the health check requests, so the health check requests do not count by the queue-proxy.

@skonto @dprotaso

Start by tracing the queue-proxy request accounting and health-check handling described in the issue. Determine how Knative Serving could distinguish model-specific health-check paths from normal requests, then define how those requests should be excluded from user metrics. The issue names no files or tests, so the implementation entry point and validation plan still need to be established.

Written by the indexing model from the issue text.