OpenAI host-compat shims

Developed by Dahl together with Gonka External TestLab.

Where it lives: the broker OpenAI HTTP proxy (POST /v1/chat/completions), before the inference gateway.

Agent → broker proxy (shims) → gateway → host validator → model

Port: this is a spec, not a library. Find the existing POST /v1/chat/completions handler; rewrite the JSON body after auth, before the gateway. Any language. Do not start a new process.