Next refinement of _M_IX86

You asked a question but closed the post for comments.

> thanks for the patch. I applied it at revision 954. Why the default is set
to 300 for x86? As I read in msdn the default should be 600, but well if it
solves your issues, I am fine. On older cpu's then i386 it is pretty
unlikely that somebody is using our CRT :)

MSDN docs say 600 is currently default value if /GB options is used.
I've taken the code from MinGW Win32 headers (windows.h) but now I can see it's broken (as if I specify -march=pentium4 then it gives only 300). And this is bad for GC lib as it uses some asm lock commands (like cmpandxchg).

On the other hand, I don't known how to fix it. The situation is follows:
__i386__ is always defined (means any 32-bit system);
__i486__ is defined if -march=i486 or no -march;
same for __i586__ and __i686__;
other pre-defined macros like __pentium__, __geode__, __k6__, __pentium4__, _nocona__, etc are defined respectively if -march is specified with the corresponding value.

So, the code should check only for __i486__ (400) and __i586__ (500) and, otherwise, gives 600. The code would be broken for i386 systems but I think it's better than be broken for modern and future processors.

The suggested patch attached.

PS. If You know the MinGW folks let them know this bug.