Upstream quota limit per worker

Your service calls a third-party payments provider whose API allows 100 requests/second across your whole account. You autoscaled your worker fleet to clear a backlog, and now a growing fraction of provider calls return HTTP 429 with Retry-After. The more workers you add, the higher the 429 rate climbs and the slower the backlog drains. Your own infra is healthy, CPU, memory, and queues on your side are fine. Dashboards show provider 429s rising in lockstep with your worker count. How do you triage and mitigate?