Book 3, 12.5; unnecessary per-term multiplication instead of single division in pdf_value of hittable_list

[Lethargy2718]

This is the current implementation of hittable_list::pdf_value in listing 56, where each object's pdf_value is multiplied by weight:

double pdf_value(const point3& origin, const vec3& direction) const override {
    auto weight = 1.0 / objects.size();
    auto sum = 0.0;

    for (const auto& object : objects)
        sum += weight * object->pdf_value(origin, direction);

    return sum;
}

However, since weight is constant across all terms, the multiplication can be factored out of the loop and applied once to the final sum instead, replacing N multiplications with a single division per call:

double pdf_value(const point3& origin, const vec3& direction) const override {
    auto sum = 0.0;

    for (const auto& object : objects)
        sum += object->pdf_value(origin, direction);

    return sum / objects.size();
}

This is a purely arithmetic change, and produces the same result, aside from possible negligible floating-point differences.

Currently, pdf_value is called on every sample and bounce of a diffuse ray when there is a hittable_list of sampled objects. For scenes with many lights/hittables, this adds a lot of unnecessary CPU cycles.

This change removes N-1 redundant multiplications per call, which leads to a shorter execution time.

It may become less readable, but that could easily be fixed by splitting it into two lines:

auto weight = 1.0 / objects.size();
return sum * weight;

I'm prepared to submit the PR for this.