[CURA-12153] Fix 'encompassing hole' issue (tree support).

[rburema]

When a hole is large enough to wholly contain (sometimes many) other branches, those also get cut. This prevents that. The solution isn't ideal, as there is an offset happening somewhere, but it's better than just missing the top of trees.

[rburema]

02 MOD.zip

Hi @ThomasRahm -- We've found a bug with this model and partly fixed it in this PR, but the fix isn't ideal because it seems to either skip or do the offset of the holes we keep in the wrong direction. Since this is a bit of an edge case, we're probably not going to return to this anytime soon, but we thought you might like to know (we don't expect anything of course, just a heads-up).

[ThomasRahm]

@rburema
I can confirm the bug this pull request fixes. But I think it was caused by git!
I vaguely remember the version with
else if (hole.intersection(PolygonUtils::clipPolygonWithAABB(relevant_forbidden, hole_aabb)).area() > hole.length() * EPSILON), but it had issues, which is why I replaced it with the slower but more accurate
else if (! hole.intersection(PolygonUtils::clipPolygonWithAABB(relevant_forbidden, hole_aabb)).offset(-config.xy_min_distance / 2).empty())
Which is present in the pull request #2088
How the former one found its way back into the code-base I do not know.
If only the latter one is used I don't have any issues, even without this workaround.

Bonus performance improvement:
If
&& ! hole.intersection(hole2).empty()) // TODO should technically be outline: Check if this is fine either way as it would save an offset
is replaced with
&& ! hole.intersection(PolygonUtils::clipPolygonWithAABB(hole2, hole_aabb)).empty()) // TODO should technically be outline: Check if this is fine either way as it would save an offset
It is a lot faster for me (15 instead of 53 seconds for the drawing part).