Drop support for clang-3.5?

Unfortunately I found out that clang-3.5 has a severe bug with the handling of std::integer_sequence which is maybe related to the sizeof... operator.

This does not only lead to compile failures. It's easy to give examples where clang-3.5 silently produces incorrect code. A dune-less test case using only std::index_sequence in an idiomatic non-artificial way for tuple-unpacking is attached to the corresponding issue in dune-functions: staging/dune-functions#10 (closed)

As a consequence we should either avoid using related features in dune or drop support for clang-3.5. Since we are already using std::integer_sequence in the core for some time I'd strongly vote for the latter to avoid people running into hard to track issues with correct code producing wrong results.