Fix tria::has_hanging_nodes for anisotropically refined meshes

[luca-heltai]

Anisotropic refined meshes do not give the correct answer for calls to has_hanging_nodes().

Working on a fix.

[peterrum]

References #13623.

[luca-heltai]

There might be a faster way, but this is only used in the anisotropic case, in serial, so for the moment, until it becomes critical in some applications, we could live with this slow fix.

[drwells]

Actually, on second thought - do we permit anisotropic refinement with p::s::T? If so we will need to do a reduction at the end.
update: I checked and the override of this function in DistributedTriangulationBase is probably wrong for p::s::T. We can deal with that later.

edit: perhaps we can cache this value in NumberCache?

[luca-heltai]

~Actually, on second thought - do we permit anisotropic refinement with p::s::T?

If it is shared, all processors will return the same value anyway.

edit: perhaps we can cache this value in NumberCache?

Yes. Maybe in a followup PR?