Performance improvements Nef_3

[GilesBathgate]

I wanted to create a discussion topic for further performance improvements to the Nef_3 package.

Previous Pull requests

• Calculate normals for facets more efficiently #5412
• Calculate indices only when required. #5438
• Bypass coplanarity assertion for triangular facets. #5455
• Explicitly call simplify after marking bounded volumes #5470
• Nef_3 performance - has_on_after_intersection in SNC_constructor #5684
• 18% performance boost in Nef_3 K3_tree. #5772
• Reserve method for Chained Map / Unique Hash Map  #6306
• Nef_3: Sphere circle constructors. #6378
• Nef_3: Use separate lists K3_tree node to avoid object casts #6480
• Nef_3: Performance std::map -> std::unordered_map #6471
• Nef_3: Avoid more filter failures in ray shoot #6470
• Nef_3: performance in binop_intersection_tests #7187

Outstanding pull requests

• Speed up transformation by calculating the transposed inverse matrix once #5590
• Nef_3: Performance SNC_external_structure Visit shell objects only once #7190
• Nef_3: Sphere map sizes #7239

Future ideas

• Can axis-aligned planes in the sideness test of K3_tree be exploited (Ortho_plane_3)
• Can CGAL_NEF3_INTERSECTION_BY_KDTREE be made to run faster. See research in https://core.ac.uk/download/pdf/47241901.pdf
• Could Intersecting Sequences of dD Iso-oriented Boxes be used for point location and ray shooting, to replace K3_tree
• dD Spatial Searching?
• AABB tree?
• Orthtree?
• No hashmap approach