AW3: improve manifoldness enforcing heuristics

[MaelRL]

Summary of Changes

There is, at the end of the alpha wrap process, a post-processing step that re-adds material to enforce geometrical manifoldness.

This step is driven by a heuristics that prioritize cells around non-manifold vertices. The current heuristic prefers:

• cells that are not incident to artificial vertices (bbox and cave initilization points)
• cells that are incident to a lot of facets
• cells that are adding a small volume

Turns out the 2nd criterion is not so great: because it ignores volume, it can add very large tetrahedra. Furthermore, it can cascade into adding absurd amounts of tetrahedra. Its purpose was to reduce the amount of tetrahedra added, hoping that by adding few tetrahedra (regardless of their volume), it would end up adding less volume than adding small tetrahedra. In practice, it seems like it doesn't work that well, so just remove it and let the process be driven entirely by tetrahedron size.

Here is a comparison for a typical case reported in #7625:

TODO:

• do not resort at every step since we don't use the #boundary facet anymore
• retest manually post bc8351f

Release Management

• Affected package(s): Alpha_wrap_3
• Issue(s) solved (if any): fix Alpha Wrapping Strange Features #7625
• Feature/Small Feature (if any): n/a
• License and copyright ownership: no change

[lrineau]

Is this PR a change that could be backported to 5.5.x or 5.6.x?

[MaelRL]

It's on the fence between bug fix and changing the behavior. I picked the latter for the (low) probability that it would be an unwanted change for some users.

[lrineau]

It's on the fence between bug fix and changing the behavior. I picked the latter for the (low) probability that it would be an unwanted change for some users.

Then, maybe rebase the branch onto a release branch, but let the PR be based on master. That will ease a possible merge of the changes by a user (by a merge of your branch).

[MaelRL]

Merged with #7805.