Surface per-model active/waiting/limit from the background task limiter so admins can distinguish task scheduling from model-service throttling, with full i18n coverage.