Currently we get a number of warnings when building the documentation for Mesh_3:
```
/home/cgal-testsuite/cgal_doc_build/CGAL-5.3.1-I-169/doc/Mesh_3/Mesh_3.txt:743: warning: unable to resolve reference to 'CGAL/Mesh_3/initialize_triangulation_from_labeled_image.h\' for \ref command
/home/cgal-testsuite/cgal_doc_build/CGAL-5.3.1-I-169/doc/Mesh_3/Mesh_3.txt:801: warning: unable to resolve reference to 'Mesh_3/random_labeled_image.h\' for \ref command
/home/cgal-testsuite/cgal_doc_build/CGAL-5.3.1-I-169/doc/Mesh_3/Mesh_3.txt:817: warning: unable to resolve reference to 'CGAL/Mesh_3/initialize_triangulation_from_gray_image.h\' for \ref command
```
Note the backslash at the end of the file name.
Most likely an attempt was made to escape the dot to not have it as part of the filename, though at least in doxygen 1.8.13 a closing dot was not seen as part of the filename.
Looking at the other escape character of doxygen: the commercial at sign, this was apparently working as escape in the filename as well, though in doxygen 1.8.15 this was not the case anymore either.
During one of the last changes also the backslash lost this possibility for this case and this is now consistent.
The current change (removing the backslashes) doesn't look like to alter the behavior for the 1.8.13 and master version in respect to CGAL.
The work was already mostly done in PR #4892, but there was still two
issues:
- `tbb::parallel_do` has been removed, and `tbb::parall_for_each`
was to be used instead,
- the support for `tbb_hasher` has been removed.
This time, I have tested in a container were
intel-oneapi-tbb-common-devel-2021.3.0-2021.3.0-511.noarch was
installed, but not tbb version 2020.
The link checker gave a number of redirects and incorrect links.
- the redirects have been solved ass far as possible
- the incorrect links have been checked and corrected where possible, others have been reported through issues;
differentiate between corner vertices and edge vertices
corner vertices are forced to be "terminal" in the call
to split_graph_into_polylines() hence leading to
as many polylines as segments on image bbox edges