Fixes #1054 by ensuring thread savety of hessian calculations with GFN-FF

[Thomas3R]

The distance array neigh%distances was not allocated thread save and was removed from the neigh class.
Now the distances are calculated from the atom coordinates and the translation vectors.

[marcelmbn]

Given the assumption that the distance array cannot be allocated thread safe, this seems like a reasonable change to me.
However, it's principle-wise not a good development to move away from modularized calculation of static variables (like a distance array) towards calculating it from scratch whenever needed.
Thus, do you think there's any chance to allocate it in a thread-safe way?

[Thomas3R]

Of course, a distance array could be used in a thread-saving way. However, I recommend implementing this together with a revision of the force field parallelization. This would be a major effort to implement either way.

[marcelmbn]

Again, assuming that there's no better solution using the distances array of the neigh type in a thread-safe manner, this seems like a reasonable change.