Add TensorProductMatrixCreator namespace

[peterrum]

I accidentally closed #14291. This contains the same content.

[peterrum]

Do you plan tests for this? I think we should move forward here, so we can do it elsewhere.

I have locally a test for this function, where I compare the results with assembled/restricted matrices; but for that I need to get some more functions into deal.II.

This looks good in my opinion. I think one question I would have for the future is whether there is a way to avoid re-computing the same reference mass and stiffness matrices, as we might call this function many times with just the cell_extent and boundary_ids differing, but the same underlying 1D matrices.

Yes, this bothers me, too. One could shift the inner part of the function into separate function that takes the two reference matrices. But in that case one needs to handle hp manually. But I guess we can do that in a follow-up PR. In general, we should probably revisit the interfaces in the next weeks/months.

[kronbichler]

The all-headers test fails. We need to include at least fe_values.h and dof_handler.h

[kronbichler]

I think this is C++17, so we need to work around it.