compute_current_constraints is called twice at the start of time step 0

[jdannberg]

At the moment, we call compute_current_constraints() twice at the start of time step 0 (and for each adaptive refinement of time step 0), once in setting the initial conditions, and then another time in start_timestep. This is not wrong, it is just a bit confusing and probably unnecessary work. So it would make sense to not call compute_current_constraints() in start_timestep if the time step number is zero. However, as @gassmoeller pointed out to me, we should first think about what happens in the case of restarting a computation and test if we run into any problems there before making this change.