fix: drop the timing from the response report log The elapsed time measured around do_socket described the timer, not the request, so it was noise in the log rather than a useful signal, and it cost two ngx.update_time calls on every reported response. Log the outcome alone.