Efficiency of TDS_3::*incident_*_threadsafe functions

[lrineau]

For the efficiency of incident_cells_threadsafe, incident_vertices_threadsafe, and so on, I think we should modify the boost::unordered_set here:

https://github.com/CGAL/cgal/blob/5a02f47c56a4d2084e20688bb2817e2017b363ef/TDS_3/include/CGAL/Triangulation_data_structure_3.h#L827-L834 and here: https://github.com/CGAL/cgal/blob/5a02f47c56a4d2084e20688bb2817e2017b363ef/TDS_3/include/CGAL/Triangulation_data_structure_3.h#L995-L1000

It should be benched, but those boost::unordered_set should probably be replaced by something more efficient:

• boost::flat_set based on small_vector (see benchmarks: that depends on the number of queries compared to the number of insertions),
• or pmr::unordered_set (from Boost>=1.60 or C++17) with a with its allocator constructed from a pmr::monotonic_buffer_resource (from Boost>=1.60 or C++17).

[vishalvibes]

@lrineau I would like to work on this.