[AD] A5 END_OF_MAIN() exorcism progress

[Evert Glebbeek <eglebbk@xxxxxxxxxx>]

My local sandbox currently doesn't need END_OF_MAIN() anymore on OS X.  
The only caveat is that for C++, the prototype has to be "int  
main(int, char**)" - unless Allegro isn't allowed to steal the main  
function, in which case the user is responsible for doing whatever  
Allegro needs the user to do in main().

In the course of the week (possibly running into next week, but  
hopefully not too much), I will:

1. Move the OS X main.m file from the main library into an addon. The  
proper way to link Allegro on OS X then becomes "link to liballegro  
and liba5-main".
2. Merge in Elias' previous patches for Windows and Linux, although  
magic main was pretty much a no-op in the latter case already and can  
be avoided in the former case as well (MinGW doesn't need it, MSVC  
doesn't need it after setting a #pragma; this should again be  
configurable of course).

At that point I will post the patch for review. I am fairly adamant in  
wanting to call the entry point in my main sourcefile "main()", no  
"al_main()" or "allegro_main()". Just "main()".
However, what I will probably do is make the code so that "main" and  
"allegro_main" mean the same thing, since having "allegro_main" as the  
entry point seems to have at least some supporters.
In cases where Allegro needs to "steal" main(), that means "#define  
main allegro_main".
In cases where Allegro doesn't need to steal main(), that means  
"#define allegro_main main".
If Allegro should not steal main() ("#define ALLEGRO_NO_MAGIC_MAIN",  
do we keep this?), neither of those is done. The entry point is called  
main() and the user is responsible for doing any initialisation that  
Allegro requires themselves and does not link to liba5-main.

Once the patch has been verified to work on all platforms and no one  
has serious complaints, I'll clean up the examples and demo program,  
document the change and commit.

Thoughts?

Evert