py26-numpy, py26-scipy, and macports atlas/lapack/blas

[James Kyle]

> I believe negative variants are still not stored in the registry, so  
> default variants can cause issues for people when upgrading a port  
> (since they'll be automatically selected again).

This was my reasoning for going with the no_* nomenclature.

>   You'll want these packages to build against the gcc44 package
> rather than the gcc43 package.

As others have mentioned, the gcc43 (and previous) packages build fine  
for me on SL (late 2008 mac pro).

> I really like this idea, but the main thing that concerns me is that
> any port that depends on numpy will now also depend on gcc43, or gcc44
> (which seems like a better idea), this is a big change as gcc takes a
> _long_ time to build.

I understand that concern. As others mentioned, I tried to include a  
way for the deps to be disabled that would be upgrade friendly as well.

The atlas port I put together conforms to the official optimization  
methods and so is quite a beast in its own right. I haven't benched,  
but I'd guess it takes almost as long as gcc.

I went with 4.3 over 4.4 as it seemed to be the consensus for a common  
compiler in ticket #20103.

I'm fine with using the 4.4. toolchain if the other packages follow  
suit. :)

-james