Split VectorTools up

[masterleinad]

Related to #9790. This also allows to only include smaller headers than all of VectorTools but I didn't make that transition for the examples or the tests.

The approach taken here, was to first split up the template file using the instantiations files and then splitting vector_tools.h matching the template files. If we don't want to split up vector_tools.h it's easy enough to revert that.

The actual changes are pretty small, it's mostly only moving code around.
Some things that changed:

• using more forward declarations
• some more explicit instantiations for dim<spacedim for the boundary functions. We were previously generating them implicitly through the project functions.

I verified manually that the documentation didn't change at first sight.

[tamiko]

I like the categorization. It is extermely hard to display the 13k line changes here, so I will trust you that they are split up appropriately :-)

[tamiko]

/rebuild

[masterleinad]

Yes, the commit history is not super clean. I can try to get every commit compiling but that's quite some more work. Otherwise, I would just squash into a few commits.

[tamiko]

Squashing at the end and having the full CI pass is all we need :-)

[masterleinad]

/rebuild

[bangerth]

Have you measured what this means for the overall CPU time? Many of us have many-core machines where this kind of change makes sense if you do -jmany. But what happens if you do -j1 or just make? Does this lead to appreciably longer compile times? Should we care about this case?

[masterleinad]

I didn't change the targets.

[bangerth]

Oh, I see -- I hadn't paid enough attention to the fact that you're not splitting up which functions are compiled in which .cc file.

[masterleinad]

Squashed.

[kronbichler]

That gets a big thumbs up 👍 by me.