Signed-off-by: Felix Schindler <felix.schindler@wwu.de>

This can be used for instance in the context of a static_assert to check
if a given function is derived from LocalizableFunctionInterface. Since we
do not only check whether the function is tagged as LocalizableFunction but
in deed derived from LocalizableFunctionInterface we need the helper class.

Thus new functions do not have to implement this.

There is finally only one FunctionProvider, all specializations are gone.
The price for this is one additional line in all functions that should be
provided which is acceptable. See the added documentation of
FunctionProvider on how to proceed for new functions. The only drawback is
that there are still four places in functions.hh that need to be touched.
This is left as an exercise for someone who fancies static loops over tuple
types (not a tuple object), I am too tired now... :)

Since we do not fully trust the vtk writer (for good reasons), creating the
directory beforehand seems like a sensible thing to do.

We fall back to write() in serial runs, avoiding the s0000-p000 prefix
in the output files. Hope this does not break anything, @renemilk?

parallel runs failed if "filename" contains directory parts since
VTKWriter wants to prefix filename with s00000$(RANK)

This reverts commit 9ab7afb6a85a422908ff6fbfa454fc140475c149.
no idea wtf I've commited here...

parallel runs failed if "filename" contains directory parts since
VTKWriter wants to prefix filename with s00000$(RANK)

Did not make much sense in the end, Rene was right after all^^

By printing the scalar frobenius norm of the matrix.