Hide the atomics into a layer of functions that can be used to update the particles atomically or not in a transparent way

accumulated from P2P, M2P, M2L and PM for each particle individually (not just
the combination of the four).

It also counts the interactions from each force contributor individually.

Simplify the calls to potential_derivatives_compute_M2P() and potential_derivatives_compute_M2L() since the inverse softening length is no longer needed.

Simplify the gravity potential calculations for M2L and M2P since the softened case should never take place any more. REplace the calculation by an error message.

Improve the M2P criterion to make sure we don't interact a particle and a multipole if the distance between them is less than the sofetning length (spline truncation radius)

Move the MPI type for multipole to a separate file. Create a bit-by-bit XOR reduction operation for multipoles. Utilise that reduction operation when exchanging the top-level multipoles.

Use temporary local copies of the D_xxx terms in the M2L_apply function to help the compiler optimize quantities.