[Feature] Leverage an (EPP managed) Kubernetes Service for InferencePool endpoint discovery?

[elevran]

What would you like to be added:
EPP currently watches all Pods via a controller runtime reconciler and then filters them based on the current InferencePool.SpecSelector.
In large clusters, this could lead to high load.
Based on discussions in #300 and #301 (and esp. these comments, it may be worthwhile to consider the use of an (EPP managed) Service that is synchronized with the InferencePool Selector and then watch its EndpointSlices.
This would delegate endpoint discovery to Kubernetes, which is potentially optimized.
A possible downside is the exposure of the Service to users via the API. The name itself can be randomized (e.g., "<inferencepool>-hash(selector)") so collisions are less likely.

Why is this needed:
Using an EndpointSlice could lead to better scaling of EPP, especially on Clusters with many Pods (and only a small fraction associated with the InferencePool).

[k8s-triage-robot]

The Kubernetes project currently lacks enough contributors to adequately respond to all issues.

This bot triages un-triaged issues according to the following rules:

• After 90d of inactivity, lifecycle/stale is applied
• After 30d of inactivity since lifecycle/stale was applied, lifecycle/rotten is applied
• After 30d of inactivity since lifecycle/rotten was applied, the issue is closed

You can:

• Mark this issue as fresh with /remove-lifecycle stale
• Close this issue with /close
• Offer to help out with Issue Triage

Please send feedback to sig-contributor-experience at kubernetes/community.

/lifecycle stale