[AD] D3D and OGL use different viewports

[SiegeLord <slabode@xxxxxxxxxx>]

(Note, these issues are reasonably important bugs but I wouldn't block 
5.1.8 on them as they've been there for years).

While fixing the half-pixel shift in the D3D code, I noticed that D3D 
and OGL backends use different viewports sizes. OGL uses the size of the 
parent bitmap, while D3D uses the size of the backing texture. If the 
bitmap has non-power of two dimensions and the system does not support 
them, the two approaches give different sizes. The projection matrix 
transforms user coordinates into the viewport coordinates, so having 
different viewports means that the same projection matrix will produce 
different outputs. Additionally, making it depend on the backing texture 
means that one needs to be very careful in using projection matrices 
with D3D.

I think the OGL method is better, as it hides the fact that there is a 
backing bitmap. I want to switch D3D to the same method, but I want to 
double check that D3D wasn't coded that way on purpose (e.g. maybe it's 
some device loss issue as always?).

Additionally... viewports provide a nice way of implementing sub-bitmap 
functionality instead of transformation hacks we do today. Neither one 
of the backends use viewports for this purpose. Was there a reason we 
implemented sub-bitmaps the way we did instead of using viewports? One 
benefit of switching to viewports is that it is one way to enable 
setting the projection matrix while drawing to a sub-bitmap. Right now 
this just plain doesn't work, and fixing it would touch about as much 
code as redoing it with viewports.

-SL