Improve Vec3.normalize()

[eikel]

Initially reported by @ClaudiusJ as #645:

inline vec3_t& normalize() {
    const value_t ls = lengthSquared();
    if(ls==0) {
        std::cerr<<"Cannot normalize a vector with zero length."<<std::endl;
        return *this;
    }else if( std::abs(ls-static_cast<value_t>(1.0)) <= epsilon ){
        return *this;
    }else{
        const value_t l = 1/std::sqrt(ls);
        vec[0] *= l;
        vec[1] *= l;
        vec[2] *= l;
        return *this;
    }
}

• Calculate sqrt only when necessary.
• Don't normalize a vector that is almost 1 (due to floating point errors). Dividing those numbers won't result in a vector that has exactly a length of 1.0 anyway. The question is: what is a good choice for epsilon? It should be chosen in a way, that a normalized vector is not touched when normalized again.

[eikel]

I would replace comparison by zero to an inequality. Instead of zero, the epsilon value could be used. Furthermore, remove the error message.

inline vec3_t& normalize() {
    const value_t ls = lengthSquared();
    if (ls > 0) {
        const value_t l = 1 / std::sqrt(ls);
        vec[0] *= l;
        vec[1] *= l;
        vec[2] *= l;
    }
    return *this;
}