Error of "znaupd: starting vector is zero, try rerunning?"

[Brch]

Issue description

When running with the Arnoldi solver in some cases, I get an error:

ERROR   | znaupd: starting vector is zero, try rerunning?
ERROR STOP 

sometimes, re-running it indeed works as suggested, but usually it does not and the problem cannot be computed at all.

Bug report

With no changes to the code, with a parameter file that reads:

&gridlist
  gridpoints = 250
  x_start = 0.0
  x_end = 0.1
/

&equilibriumlist
  equilibrium_type = "rayleigh_taylor"
/

&solvelist
  solver = "arnoldi"
  arpack_mode = "general"
  number_of_eigenvalues = 10
  which_eigenvalues = "LI"
  maxiter = 16000
  ncv = 400
/

&savelist
  write_eigenfunctions = .true.
  show_results = .true.
/
&paramlist
  k2 = 1
  k3 = 0
/

Actual result

  INFO    | the state vector is set to [rho, v1, v2, v3, T, a1, a2, a3]
  WARNING | overriding x_end with 0.10000000
  WARNING | standard equilibrium conditions not satisfied!
            location of largest discrepancy (1): x = 0.02396385
            value of largest discrepancy (1): 0.13546941E-11
            amount of nodes not satisfying criterion (1): 984
 
  INFO    | ---------------------------------------------
                          << Grid settings >>
            geometry           : Cartesian
            grid start         : 0.00000000
            grid end           : 0.10000000
            gridpoints (base)  : 250
            gridpoints (Gauss) : 996
            gridpoints (matrix): 4000
                      << Equilibrium settings >>
            equilibrium    : rayleigh_taylor
            wave number k2 : 1.00000000
            wave number k3 : 0.00000000
            default params : True
                        << Physics settings >>
            flow               : True
            external gravity   : True
            radiative cooling  : False
            thermal conduction : False
            resistivity        : False
            viscosity          : False
            hall mhd           : False
                        << Solver settings >>
            solver      : arnoldi
            ARPACK mode : general
            number of eigenvalues : 10
            which eigenvalues     : LI
                        << DataIO settings >>
            datfile name         : datfile
            output folder        : output
            write matrices       : False
            write eigenfunctions : True
            write derived eigenfunctions : False
            write eigenfunction subset : False
            ---------------------------------------------
 
  INFO    | solving eigenvalue problem...
  WARNING | Arnoldi: maxiter (16000) below recommended 10*N (40000)
  ERROR   | znaupd: starting vector is zero, try rerunning?
ERROR STOP 

[n-claes]

This problem seems to be an issue with ARPACK itself and not on our end, for some reason the forcing of the starting vector into the range of OP returns zero. I think I found a workaround though, in the referenced commit above I modified the ARPACK configuration to use a manually initialised starting vector using a (-1, 1) random uniform distribution and that seems to solve (all) issues. I've also seen this problem popping up in our unit tests every so often, that seems to be fixed as well now.

The fix will soon be available on the develop branch.