This combinatorial choice seemed like a good idea, but it can have
nasty cascading effects, adding very large tetrahedra. See this
issue: CGAL#7625

In the end, the only thing we care about is small volumes being added.

I keep the artificial vertex for now, but I am not fully convinced
these should be actually kept too.

Thanks @albert-github!

In a normal run of the algorithm, we shall never ask the facet status
of a facet that is already outside, but it's better to be complete
and it costs nothing.

…ormal facets

Artificial facets are *not* infinite facets.

…tion

Meaning, use the value that we compare against alpha, and not
simply the radius of the smallest circumscribing ball.

This strongly changes the order of the queue and thus thus results
are very different, but still the same (same guarantees, same
element quality, only a little bit more elements, etc.)

Also a massive, ~35% speed-up, that needs to be investigated.

This doesn't bring any speed-up because it was a very fast exit
in push_facet(): the neighbor was necessarily outside (since we
come from it), and we are done.

…ness

If one day this becomes annoying because one wishes to call
oracle.add_XXX() multiple times AND it's a significant
runtime burden, we can just add a function add_XXXs()
with a single call of accelerate_distance_queries()

There was a need for sorting at every iteration when the sorting
used criteria which were changing with every iteration. This
is no longer the case after c7b9317.

Also make it deterministic.

But keep the old one, because we might need something sorted in the future.

Also it is needed to do interrupting, because the intermediate is really
ugly for a LIFO queue.

…ness

So it doesn't purge zombies or print gate info that would not be refined

…nt-GF' into AW3-Resume_aw3-GF

Otherwise, it creates holes if we reset manifold flags...

Neither are able to produce a closed, combinatorial manifold surface
100% of the time, so minimize the diff.

Makes it easier to detect faces that no longer need
to be treated (zombies).

For outward wrapping

…const&

Reasoning is: this is what users expect.