Fix FindMKL, search for clang-format-3.9

Some cmake changes

Some fixes for new PerThreadValue for clang and dune-gdt

- see https://gcc.gnu.org/gcc-6/porting_to.html#overloaded-abs

Rebase PerThreadValue on tbb::enumerable_thread_specific

The previous implemenation had the problem that tbb does not guarantee a fixed maximal number of threads even if it is set in task_scheduler_init.

Properly fix ambiguous operator* for FieldMatrix