Support non-blocking i/o across platforms

This change introduces new tests for `O_NONBLOCK` and `SOCK_NONBLOCK` to
confirm that non-blocking i/o is now working on all supported platforms,
including Windows. For example, you can now say on Windows, MacOS, etc.:

    socket(AF_INET, SOCK_STREAM | SOCK_NONBLOCK, IPPROTO_TCP);

To create a non-blocking IPv4 TCP socket. Or you can enable non-blocking
i/o on an existing socket / pipe / etc. file descriptor by calling fcntl

    fcntl(fd, F_SETFL, fcntl(fd, F_GETFL) | O_NONBLOCK);

This functionality is polyfilled on older Linux kernels too, e.g. RHEL5.
Now that fcntl() support is much better the FIOCLEX / FIONCLEX polyfills
for ioctl() have been removed since they're ugly non-POSIX diameond APIs

This change fixes a weakness in kprintf() that was causing Windows trace
tools to frequently crash.

libc/sock/fixupnewsockfd.c

@@ -21,11 +21,6 @@
 #include "libc/sysv/consts/o.h"
 #include "libc/sysv/consts/sock.h"
 
-/**
- * Applies non-atomic file descriptor fixups on XNU or ancient Linux.
- *
- * @param fd of -1 means no-op
- */
 int __fixupnewsockfd(int fd, int flags) {
   return __fixupnewfd(fd, (((flags & SOCK_CLOEXEC) ? O_CLOEXEC : 0) |
                            ((flags & SOCK_NONBLOCK) ? O_NONBLOCK : 0)));