Improve robustness of SignedDistance computation (redux)

[kennyweiss]

We improved the robustness of the computeSign() functionality of quest's SignedDistance query in #706, but found a new set of regressions in regions with sharp angles and poorly shaped triangles (it is not clear if the problem is due to very acute angles or very obtuse angles, but both are present in the regression).

I ran the signed distance query on the following simple mesh comprised of 24 triangles

using the following command line:

> ./examples/quest_signed_distance_interface_ex -f ../data/quest/fin0.stl \
                                                --box-dims 33 33 33       \
                                                --box-min -10 -20 5       \
                                                --box-max 10 0 25

(note: fin0.stl is not yet checked into our data repository)

The computed signed distance has several regions where it computes the wrong sign, e.g. it marks points as "INSIDE" when they should be marked as "OUTSIDE". This is demonstrated in the following visualization of the zero-level isosurface of the computed distance function, which should not have any isosurface components outside the shape:

[kennyweiss]

The problem was related to two conditions:
(a) An edge with an acute dihedral angle. This creates a larger "region of influence" where the edge is the closest point on the surface to the query point
(b) A large difference in the areas of the two triangle. This effectively weights the normal towards the larger triangle, and creates "pockets" within the "region of influence" that can get the incorrect sign.

It is fixed in #784 by unitizing the normals before adding them, which removes the term associated w/ the triangle areas.