Unify lagrange cube implementations

With the recent reimplementation of Q1LocalFiniteElement using
the new LagrangeCubeLocalFiniteElement, the unit test program 'test-q1.cc'
cannot be compiled with gcc-5 or gcc-6: The compiler simply starts
grabbing more and more memory, until the machine stalls.

This patch brings back a part of the old Q1LocalFiniteElement
implementation for these cases.

For some reason clang-3.8 requires this.

This new class is intended to unify the existing implementations for
Lagrange elements on cube domains with compile-time dimension and
order.  This will make the API simpler (less implementations classes),
and the code gets shorter, too.

This commit also replaces the existing Q1LocalFiniteElement
implementation by the new one based on LagrangeCubeLocalFiniteElement.
This happens in a completely backward-compatible manner.