`LIFORequestQueue`: compute retry-after based on observed processing rate

[tgeoghegan]

In #4050, we add a response header Retry-After: 30 if the queue is full or if a request times out without being serviced. We're guessing that 30s later is probably a better time. It would be if the request queue kept track of how long it takes for dequeued items to be processed (on average? P99? not sure) and used that value to provide guidance to clients on when to retry.