prov/tcp: flush potential wait obj interupts from uring on close

When using a FI_WAIT_FD wait object with uring polled connections
it's possible for interrupts to be sent from the kernel after the
urings are unregistered.  To circumvent potential "dangling" interrupts
later in the applcation flow, do one final epoll after destroying urings

this issue was caught by the eq_wait_fd_sread flow of fi_eq_test

prov/tcp/src/xnet_progress.c

@@ -1951,6 +1951,12 @@ void xnet_close_progress(struct xnet_progress *progress)
 		free(progress->cqes);
 		xnet_destroy_uring(&progress->rx_uring, &progress->epoll_fd);
 		xnet_destroy_uring(&progress->tx_uring, &progress->epoll_fd);
+
+		/* destroying a uring may trigger an interrupt on subsequent
+		 * epoll_wait so force catch the interrupt and ignore it
+		 */
+		(void) ofi_dynpoll_wait(&progress->epoll_fd, &progress->events[0],
+					ARRAY_SIZE(progress->events), 100);
 	}
 	ofi_dynpoll_close(&progress->epoll_fd);
 	ofi_bufpool_destroy(progress->xfer_pool);