What would you like to be added:
An InferenceModel controller is a separate deployable from the EPP

Why is this needed:
Gateway controllers will only run reconciliation on the InferencePool object, and set the status, as the EPP is a part of the InferencePool.

The EPP will, in the long term, be HA, and likely active-active, so should remain read-only. We need a distinct deployable controller that can handle validation and status updates of the InferenceModel