Correct integration in time of the mesh forces

Complete rewrite of the time integration of the gravity mesh forces. This fixes the accuracy issues seen in the large-scale power-spectrum in cosmological runs.

This includes:

• Removing the mesh_force task.
• Introduce a global time-step for the mesh calculation.
• Compute the mesh forces when that time-step happens and not when we rebuild.
• Rewrite the operations in the kick tasks to simplify the with/without cosmology logic.
• Add the mesh accelerations to the gpart, remove the gravity acceleration from the xpart.
• Remove the alignment constraint on the gparts to reduce the memory footprint.
• Add an option to not compute the potentials.
• Remove the option to dither the particles at tree rebuilds. This was not the solution to the problem and was not implemented in an efficient way.

Todo:

• Clean-up the mesh time-step calculation.
• Deal with exact gravity force checks.
• Interpolation of the hydro velocities in snapshot.
• Deal with the time-step limiter.
• Deal with the time-step synchronization.
• Deal with the potential.
• Deal with star formation.
• Fix statistics.
• Documentation of the whole gravity scheme parameters.