Raise inference capacity one gate at a time

For an inference service with gateway admission limits ahead of engine limits, change capacity sequentially: reproduce a baseline, raise offered concurrency, then gateway slots, then engine sequences, and only then any serialized preprocessing or turn-detection concurrency.

At each step, classify errors and enforce tail-latency and error-rate guardrails. A staged benchmark does not prove production-safe capacity, and gateway configuration, engine capacity, and observed workload demand remain separate gates.