Inside the ModelPlane routing engine

How an LLM request flows from auth to upstream inference: config resolution, target trees, retries, and accounting kept off the hot path.