Update particles output to limite the number of processes involved -- manage a minimum chunk size -- write a multiple of a chunk

Make it works when the number of processes is not a factor of the size of the problem -- FFTW will give a partition of size 0 to several processes at the end in such case

Add an interface builder (with runtime values for spline mode and interpolation size) -- templatize the real number types (splines still rely on double)