Use scalapack or elpa for solving the eigenproblem in the Casida run
mode. For this, the whole parallelization is adapted to work both with
and without scalapack/elpa.

To use this feature, set CasidaUseScalapackLayout to true. With
CasidaParallelEigensolver, one can control if the eigensolver from
Scalapack or from ELPA should be used.

Solving the eigenproblem in parallel is of advantage if the matrix size
becomes larger than a few thousand. This parallelization can be used
together with domain parallelization.

Tests for this feature are added in the testsuite.