no point adding altivec now that apple has gone intel, so lets spend the
time optimizing for SSE2

we should use NASM for the assembler, and add a #define
PIXELTOASTER_USE_SSE2 which interfaces to the external assembly routines.

initially we should focus on the key conversion routines, eg. floating
point color -> integer, as these are pretty slow at the moment, as they are
pure c++ (however, they are as fast as possible for c++ impl!)

next, we should optimize common cases of 32bit -> 24bit/16bit etc.