1) Flipped the order of "discover_vertical_shells" and "process_external_surfaces",
now the external surfaces are expanded after "discover_vertical_shells"
aka "ensure vertical wall thickness" is solved.
2) Reworked LayerRegion::process_external_surfaces() to only expand into
"ensure vertical wall thickness" regions, also the expansion is done
in small steps to avoid overflowing into neighbor regions.
also:
Utility functions reserve_more(), reserve_power_of_2(), reserve_more_power_of_2()
Various SurfaceCollecion::filter_xxx() modified to accept an initializer list of surface types.
New bridges detector refactored to accept overhang boundaries.
BoundingBoxWrapper was moved from RetractCrossingPerimeters to AABBTreeIndirect.
1) Reworked the merging code to use an AABB tree for better locality.
The old code sorted lexicographically, the new code splits bounding
boxes by the longest axis.
2) Refactored to a functional style with better const correctness.
3) Reduced memory allocation pressure by replacing std::set with
vectors, in place merging etc.
Small refactoring of AABB tree distance query function, to allow different primitives (apart from triangles)
Implemented Distancer and functions to create AABB tree from lines and use closest point query
Added test for the AABBTree with lines
Added Benchmark comparing EdgeGrid with AABBTree on line contours (Inside AABBTree test file, disabled under compilation flag)
Replaced shared_ptr<> with deque.
Merged multiple vectors into one.
Refactoring using common helper functions (prev/next_idx_modulo(),
angle(), ...)
AABBTreeIndirect::intersect_ray_all_hits(): Reuse memory of the hits cache.
The triangle-ray intersection function used a hard coded epsilon,
which did not work for triangle meshes, that were either too small
or too large. Newly the epsilon may be provided to the AABBTreeIndirect
search functions externally and IndexedMesh calculates a suitable
epsilon on demand from an average triangle mesh edge length.
Bounding boxes of polygons could overlap. Ask the AABB tree for all possible candidates.
Might be faster than searching for the closest triangle, that requires traversing the whole depth of the tree every time.
1) Octree is built directly from the triangle mesh by checking
overlap of a triangle with an octree cell. This shall produce
a tighter octree with less dense cells.
2) The same method is used for both the adaptive / support cubic infill,
where for the support cubic infill the non-overhang triangles are
ignored.
The AABB tree is no more used.
3) Optimized extraction of continuous infill lines in O(1) instead of O(n^2)
