- remove redundant loop when reuniting outputs from all threads

- For safety, since some loops are murky, overallocate some arrays of size nspheres instead of configurations
- correct a race in the definition of myompthread
- add some omp barriers to force synchronisations

Move the content of the loop over scales in a separate function, so it can be called with fresh arguments every time to check parallelism and dependencies.

- turn C4 from a structure to a class (to do the above), move its initialisation to its constructor
- eliminate "allocate_vectors()", redundant and including a (wrong) double allocation