Improve Mesh_3 memory footprint

[MaelRL]

heaptrack | vtune | usr/bin/time

• mesh vertex base: number_of_incident_facets_ and number_of_components_ to int (or even smaller)?
• get rid of sliver_value and use a local unordered_map<Cell_handle, sliver_value> in the optimizers?
• get rid of subdomain index? ==> could be interesting because not all cells are in the domain, due to alignment, we gain nothing at the moment
• canonical facets in the C3T3 map (more complex than it seems: the canonical facet could be in the conflict zone, but the information shouldn't be lost when it's removed from the complex, before the creation of a new facet)
• can gain a little bit with writing a compact_mesh_vertex_3: there is a for example a Boolean in the TDS vertex base
• Reduce the size of Point_3*? In sequential it's straightforward (e.g. quietNaN), but concurrent? But it's only stored for cells in complex, so is it worth it (3 doubles always vs 1 always and 1+3 often)
• MISC: Can have memory leaks in parallel if multiple new circumcenters at the same time?

TODO:

• Announce breaking change on criteria template parameter
• Fix binary I/O (SMDS_3 - C3T3 I/O #9376)

[sloriot]

MSVC2017 has a compilation issue:
https://cgal.geometryfactory.com/CGAL/testsuite/CGAL-6.2-Ic-157/Generator/TestReport_Christo_MSVC2017-Release-64bits.gz

 C:\CGAL_ROOT\CGAL-6.2-Ic-157\include\CGAL/Mesh_3/sliver_criteria.h(162): error C2327: 'CGAL::Mesh_3::Sliver_criterion<Tr,Cache,CGAL::Mesh_3::Sliver_caching_policy::BELOW_BOUND,std::vector<CGAL::internal::CC_iterator<CGAL::Compact_container<CGAL::Compact_mesh_cell_3<CGAL::Point_3<Kernel_>,CGAL::Weighted_point_3<Kernel_>,Subdomain_index,Surface_patch_index,Index,TDS2>,CGAL::Default,CGAL::Default,CGAL::Default>,false>,std::allocator<_Ty>>>::caching_policy': is not a type name, static, or enumerator [C:\CGAL_ROOT\CGAL-6.2-Ic-157\cmake\platforms\MSVC2017-Release-64bits\test\Generator\generic_random_test.vcxproj]

[sloriot]

Testsuite looks good in Successfully tested in CGAL-6.2-Ic-160