Add /v1/systemone and /v1/decisions simulation for vLLM #786

What would you like to be added?

Support these non-streaming endpoints for vLLM, using its request and response formats:

• POST /v1/systemone
• POST /v1/decisions

Return deterministic simulated results and reuse the existing latency simulation, load accounting, and failure injection. Include protocol tests and document that scores and usage are synthetic.

Why is this needed?

Enable CPU-only routing and failure tests for Decision API requests without loading real models. Real-engine scoring and compatibility validation remain out of scope.