Why does the specular part of the Cook Torrance BRDF model,
D*G*F/(V.N*L.N) divide by V.N (cosine of the view angle) as well as
the L.N (cosine of the incident angle)? How does this make sense in
terms of a set of flat microfacets where the reflectivity (L.N*brdf)
should just be the Fresnel term?

(ps. I have read Cook and Torrance's paper and Torrance and Sparrows
earlier paper and still don't understand)