Improve parallelization of the hydrogen bond potential in GFN-FF calculations

[Thomas3R]

Hydrogen bonds (HB) are present in almost every GFN-FF calculation, and the number of HB interactions often becomes very large very quickly (especially for periodic systems).
The bottleneck are the gfnff_hbset and gfnff_hbset0 calls, which are potentially called in every energy+gradient calculation of the force field. The HB-list setup in both subroutines is not parallelized and seems to create an overhead (eventually due to bad compiler flags).

Implementing OpenMP parallel loops for the mentioned subroutines should significantly improve the scalability of the force field (for HB-dominated systems).