Major overhaul of the x86 assembly for the decoder.

Reduced the divergence between x86-32 and x86-64 (there is now basically 
 one function that is different between them, and that due to the extra 
 registers x86-64 gives us).
Used named parameters for readability (requires gcc 3.1 or later).
All global symbol references should now be handled by gcc, so there 
 should be no problems with/without -fPIC, or various underscore 
 conventions for different toolchains (e.g., OS X).
Currently satisfies register allocations even WITH -fPIC and WITHOUT 
 -fomit-frame-pointer (the worst case scenario).

The same cannot be said for the encoder's asm, on basically any of these 
 points.

svn path=/trunk/theora/; revision=12873