Specialize integer implementation of YCoCg-R to use bitshifts

[Mokosha]

It turns out that the current implementation of YCoCgR conversion is not reversible (f(g(x)) != x) for certain combinations of pixels, for example:

glm::ivec3 v(226, 124, 192);
// This fires, when it shouldn't if the implementations are reversible.
assert (v == YCoCgR2rgb(rgb2YCoCgR(v)));

I've added a patch that specializes the template instantiation for integer types to use bit shifts as outlined in Eqn (8) of this paper. This fixes the situation in the assert, yet leaves floating point implementations to work as they used to.

[Groovounet]

Thanks for contributing!
Christophe