This test was wrong, it was computing A * A^T instead of A^T * A.
The latter is correct since our implementation of Cholesky decomposition
stores the triangular matrix in the upper triangle
The test was passing only because the Cholesky decomposition was tested
with diagonal matrix only, than this does not matter.

plot.py adapted to automatically create some graphs
more fixes in results_sorted.txt for scalapack_all when failed
some plot pdfs stored. Not sure about this...

by inspecting the 2 respective log files, it is clear, that sclaapack
failed here, but the timer still provided a result:

{    7,    3}:  On entry to·
PSSYEVD parameter number   14 had an illegal value·
{    0,    0}:  On entry to·
PSSYEVD parameter number   14 had an illegal value·
 /= Group                                              [s]     fraction
 |                                            ============ ============
 |_ e%eigenvectors()                              0.001188        1.000
{    1,    0}:  On entry to·
PSSYEVD parameter number   14 had an illegal value·
{    2,    0}:  On entry to·
PSSYEVD parameter number   14 had an illegal value·

thus those 2 values were discarded

to work with previously provided results file

I also included text file, which is the product of this processing for
the scaling performed this summer and can be used to retrieve all the
data needed for plots. It also contains references to original
log-files, so it is easy to find where is each number going from. Of
course, one needs the original raw data (log files), which are at the
moment only on my hydra account. Maybe we should put them somewhere more
visible?

This should improve performance, we will see on timings
The main problem was in normalizing eigenvectors, now we
compute maximum of computed (not analytical) solution.
It was nicer to find maximum of analytical one, since it
is the reference, but the implementation was really slow

to tell what is making the test so slow

matrix was initialized outside of three diagonals

in Fortran test.F90

both fortran and C tests generated in generate_automake_test_programs.py
linker flags added by hand -- FIX this

matrix for the test not prepared (called with zero matrix)
list of gitlab tests temporarily reduced

in elpa2_bandred_template.F90

in elpa2_bandred_template

putting scripts on one place in the repository in an attempt to make
running and processing scaling tests more easy and systematic
Scripts will be further improved

But allow this in cholesky and hermitian multiply test