numerically unstable zonal flows when maxwellian

[DenSto]

This is a similar issue to #59, but seems to be much stronger and present even when zed_upwind = 0.

When doing the standard RH test case with adiabatic electrons and maxwellian_inside_zed = .true., thing act as they should. However, if instead one uses kinetic electrons (nspec =2), the ky = 0 mode become wildly unstable. I'm not quite sure why this is, and I think it would be good to have this case working, since I think how one should center gradpar and maxwell_mu is a little more clear in this case.