The way the implementation works today is that a user label a gateway with istio.io/enable-inference-extproc=true, and this attached a listener-level extProc filter on disabled mode, but will some relevant configs that we need, and that are not available in ExtProcPerRoute proto.

Then when we encounter a ref to an inferencePool from a route, we add ExtProcPerRoute which sets the destination of the extension and override it to enabled for this route.

We need to decide on:

• Do we want to retain this logic of listener level config in disabled mode, and overrides per route?
• Do we want to leave this label-based opt-int? (I vote no, it was just an additional guard on top of the flag-based-enablement and it provide no value)