Cost-governed LLM execution for Laravel: per-profile spend ceilings enforced before the call, model fallback, JSON Schema validation with one re-ask, and real cost accounting per run.