Optimize receive_msg_core() to a single walk over the entries #1603

receive_msg_core() walks the entries twice: once backward to find where the subscriber's qos_depth deliverable messages start, once forward to hand them over oldest-first, so every entry in the range goes through find_publisher_info() twice. A single backward walk that keeps the last MAX_RECEIVE_NUM entries it accepts in a ring would cover it in one pass, at the cost of rewriting the delivery loop. Alternatively the ioctl could return the entries newest-first and let agnocastlib read the array in reverse, which is simpler still but changes the kmod/agnocastlib contract and has to keep ret_call_again able to walk a window wider than MAX_RECEIVE_NUM.

The second walk came in with #1602, where the smaller diff was preferred with a release close.