Empty level fix

[sw561]

Summary

Closes #3516 . This works for me in practice, and is more convenient than filtering the levels on the caller's side.

Additional background

Idea is to simply not consider the empty levels in the linear solver.

Checklist

The proposed changes:

• fix a bug or incorrect behavior in AMReX
• add new capabilities to AMReX
• changes answers in the test suite to more than roundoff level
• are likely to significantly affect the results of downstream AMReX users
• include documentation in the code and/or rst files, if appropriate

[WeiqunZhang]

One is expected to use the following AmrMesh functions to get objects for the linear solver.

Vector<Geometry> Geom (int a_coarsest_lev, int a_finest_lev) const
Vector<BoxArray> boxArray (int a_coarsest_lev, int a_finest_lev) const
Vector<DistributionMapping> DistributionMap (int a_coarsest_lev, int a_finest_lev) const

And if the coarsest level for the solver is not AMR level 0, the versions without arguments that get the objects starting from AMR level 0 will always be wrong.

I am not sure if we should try to guess the user's intention or add an assertion making sure BoxArrays are not empty. Let me think about this.