tcp net subsystem: handle data race gracefully

It may happen that a socket file descriptor has been closed either
while setting up poll() et al or while being inside the system call.
This was previously treated as error and caused abort in debug
builds. However, it was essentially ignored in production builds.

This has now been fixed and now is always gracefully ignored. This
most importantly fixes some flakes in CI runs (which were caused
by this situation).

runtime/nsdsel_ptcp.c

@@ -159,9 +159,9 @@ IsReady(nsdsel_t *const pNsdsel, nsd_t *const pNsd, const nsdsel_waitOp_t waitOp
 
 	const short revent = pThis->fds[idx].revents;
 	if (revent & POLLNVAL) {
-		DBGPRINTF("ndssel_ptcp: revent & POLLNVAL is TRUE, something is wrong, revent = %d", revent);
+		DBGPRINTF("ndssel_ptcp: revent & POLLNVAL is TRUE, we had a race, ignoring, revent = %d", revent);
+		*pbIsReady = 0;
 	}
-	assert(!(revent & POLLNVAL));
 	switch(waitOp) {
 		case NSDSEL_RD:
 			*pbIsReady = revent & POLLIN;