Correction of the computation of the energy spectrum

[QuentinMale]

As discussed here: #153, here is a correction for the energy spectrum calculation.

The ParallelFor now loops on 0,1,2,...,nx 0,1,2,...,ny and 0,1,2,...,nz with a change of i index if i > nx/2 to take into account the Hermitian symmetry of the R2C output array.

I verified that the output spectrum is correct by comparing with C2C which outputs the full array.

[WeiqunZhang]

The changes will not work on multiple processes because the domain might be decomposed in x-direction. What we have to do it for i > 0 && i < (N+1)/2, we double the value.

[WeiqunZhang]

I think we need a special treatment for N/2 too when N is even. My original suggestion i < (N+1)/2 is cryptic. A better way is 2*i != N.