[libcvd-members] SSE functions -- RFC

[Edward Rosten]

Here's a question of style:

I have a straightforward implementation of tensor voting, which performs 
all the maths in double precision Matix<2>.

I also have a version which performs the maths with single precision SSE, 
and is about 3x faster in some cases. They're not really quite 
interchangable because of the lack of precision.

What should I do with it? It's checked in, but you have to define 
CVD_EXPERIMENTAL to access it.

Comments/suggestions?


-Ed