MPI calculation fails on Polaris when split over 2 nodes with fluid (nan Energies)

[ColinBundschu]

Complete input and outputs here: test.zip

It seems like there is an issue with the MPI encoding/decoding leading to nan

# Energy components:
   A_diel =                      -nan
   Eewald =     3622.1040525149132918
       EH =                      -nan
     Eloc =                      -nan
      Enl =                      -nan
      Exc =                      -nan
 Exc_core =        5.2485462687340156
       KE =                      -nan
        U =                      -nan
-------------------------------------
     Etot =                      -nan
       TS =        0.3917036407203660
-------------------------------------
        F =                      -nan