Modify glm::refract's behaviour

[tigert1998]

Refer to issue #806.

No offence, I don't think the changes in 76580af is appropriate. Actually, eta's absolute value can be greater than 1.

When a bundle of light shoots from a denser medium (say, water whose refractive index equals 1.3325) to less dense medium (air, for instance), a full internal reflection is possible to happen.

In this case, eta is 1.3325 / 1 = 1.3325. If we constrict eta into the range from -1 to 1, glm will never support this kind of phenomenon, which is quite significant in computer graphics.

By the way, negative eta may be redundant.

[Groovounet]

You are quite right my change didn't made much sense.

Thanks for contributing!