The were a couple of places where a load of a stack variable was too
wide. Detected with the

  -fsanitize=address

flag of GCC, that we should probably incorporate into our CI tests.

Anything if it makes Andreas happy :)

The timer object was not properly removed when elpa_destroy() was called
on an ELPA instance, leading to a memory leak

Previously you could set up ELPA with nev > na and force a crash.

even if only part of eigenvectors are calculated, all eigenvalues should
be checked

We got reports from a user that there were troubles with certain domain
decompositions. So far the tests only looked at (approximately) square
decompositions in column-major process order.

Now, a new class of tests loops over all possible decompositions
(row * col) for a given number of total tasks.

So far, we can not confirm that there are any problems, all
possibilities work as expected.

matrices of dimension of the form 2^n * 3^m * 5^m are now allowed.
For other matrix sizes, the test is terminated. Matrix size is not
modified, as it has been the case before

There was again a case where stack-variables were loaded with
instructions that needed properly aligned memory. This only surfaced
with the Intel C compiler, where the stack layout evidently was
sufficiently different to trigger this.

This was also the case for SSE kernels

- remove runtime by not executing unecessary runs
- make sure that on gp02 impi/2017 is NOT use
- adjust reserved time on knl systems