Add FFTW support for Poisson solver and optimize distribution handling

Add FFTW support for Poisson solver and optimize distribution handling

• Swap the ordering of MPI ranks in directions Y and Z in the UNIFORM distribution used for the FFT. Now Y is the fast direction, leading to a more natural X, Y, Z ordering that has benefits for IO.

• A communication phase is saved in the 3-D FFT in the Poisson solver by not going back immediately to the canonical x-pencil distribution after the 'forward' operation.

• Added support for the use of FFTW for the one-dimensional ffts. If the FFTW library is compiled in, this feature is controlled by the option

      Poisson.Method [FFTW | Internal]

  where 'Internal' refers to the built-in code.

Author: Rogeli Grima (BSC)

Closes #169 (closed)

Closes #170 (closed)