Webhook delivery queue isolation

Design the ingress queueing and shaping for a webhook-delivery system that must deliver events to thousands of customer endpoints with wildly different processing speeds, some endpoints respond in 50ms, others in 10s, and some are down for hours. A naive shared queue lets slow/dead endpoints clog delivery for fast ones (head-of-line blocking across customers). Design the queueing model, per-endpoint shaping and backoff, how you isolate slow consumers, and how you bound memory/queue growth when an endpoint is down for a long time.