Since our unix receive queue got longer, the test run-t_sendrecv

Since our unix receive queue got longer, the test run-t_sendrecv
did not terminate anymore on some machines.  The test counts 100
send errors before it finishes.  NetBSD has added sched_yield() in
the receiver loop to trigger the errors on the sender side.  Although
not perfect, it works for me.  Get current t_sendrecv.c from NetBSD.