The patch seems obviously correct to me; there's no way a user of pythread.h can make NO_EXIT_PROG be undefined.  The patch no long applies cleanly to thread_nt.h -- one hunk is rejected -- but it looks like the change in that hunk has already been applied (though I can't try compiling on Windows to verify that everything still works).  I did try applying the patch and re-compiling on Linux.

I think you can just apply it; if you're cautious, you could ask someone to try it on Windows first.