Shared memory parallelization for the bond loop

[hidekb]

Description of changes:

• Convert arrays of structs (AoS) to structs of arrays (SoA) to improve the performance of the bond loop.
• The bond loop is decomposed into three distinct components, namely pair bond, angular relationships, and dihedral configurations.

[RudolfWeeber]

Given the rather large number of member variables this adds to the CellStructure, would it be possible to move all the bond list managment business inclduing the clearing/allocating functions into its own class, say LocalBondState and only hold the unique pointer to that in the CellStructure?

[hidekb]

What ws the reason for the additoin of kvisc=[nearly ero] in the oif_volume_conservation test?

I modified the implementation within the conditional branch if (kvisc > tiny_oif_elasticity_coefficient), and the accompanying test script must execute this path to satisfy coverage requirements. The comment within this branch indicates that it is not fully implemented. Hence, we may disable it entirely by commenting it out. At the time, I used a very small value of kvisc for preliminary testing.

[hidekb]

Given the rather large number of member variables this adds to the CellStructure, would it be possible to move all the bond list managment business inclduing the clearing/allocating functions into its own class, say LocalBondState and only hold the unique pointer to that in the CellStructure?

I refactored the code. Now, variable- and function-related bond storage has been moved from CellStructure to LocalBondState.

[jngrad]

Node-level performance improvement: a LJ simulation with harmonic bonds is now faster with 4 OpenMP threads and 1 MPI rank vs. 1 OpenMP thread and 4 MPI ranks.