Make the package installable

[mortele]

• Add a pyproject.toml file in the base directory with [build-system] specifications to avoid a bug currently in pmesh, mp-sort, and pfft-python which causes pip install to fail unless Cython, numpy, and mpi4py is installed before pip install is called. C.f. PEP 518.
• Write setup.py
  • Handle compilation of the fortran files as distutils.extensions. If this turns out to be too much of a hassle, we might have to rewrite the bond calculations in C and use Cythonize.Build
  • Fix an automatic versioning scheme derived from the commit hash. I really cba manually handling minor version numbers.

[mortele]

Done (except for the last parts, which we postpone for now) in e411b90

[mortele]

We need to reopen work on point 2.1, the memory layout changes necessary between bonded and non-bonded force calculations, because of the Fortran implementation, is not stable enough over long simulations. We should stop postponing this and transition finally to a Cython or pure C implementation.

[mortele]

After this is finished we can remove all of the following:

HyMD/hymd/main.py

Line 294 in ec41612

bonds_4_coeff = np.asfortranarray(bonds_4_coeff)

HyMD/hymd/main.py

Lines 338 to 344 in ec41612

	positions = np.asfortranarray(positions)
	velocities = np.asfortranarray(velocities)
	bond_forces = np.asfortranarray(bond_forces)
	angle_forces = np.asfortranarray(angle_forces)
	dihedral_forces = np.asfortranarray(dihedral_forces)
	dipole_positions = np.asfortranarray(dipole_positions)
	transfer_matrices = np.asfortranarray(transfer_matrices)

HyMD/hymd/main.py

Line 384 in ec41612

dipole_positions = np.asfortranarray(dipole_positions)

HyMD/hymd/main.py

Lines 650 to 653 in ec41612

	positions = np.asfortranarray(positions)
	bond_forces = np.asfortranarray(bond_forces)
	angle_forces = np.asfortranarray(angle_forces)
	dihedral_forces = np.asfortranarray(dihedral_forces)

HyMD/hymd/main.py

Line 675 in ec41612

bonds_4_coeff = np.asfortranarray(bonds_4_coeff)

HyMD/hymd/main.py

Lines 703 to 704 in ec41612

	dipole_positions = np.asfortranarray(dipole_positions)
	transfer_matrices = np.asfortranarray(transfer_matrices) # noqa: E501

HyMD/hymd/main.py

Lines 627 to 630 in ec41612

	positions = np.ascontiguousarray(positions)
	bond_forces = np.ascontiguousarray(bond_forces)
	angle_forces = np.ascontiguousarray(angle_forces)
	dihedral_forces = np.ascontiguousarray(dihedral_forces)

[Lun4m]

We need to reopen work on point 2.1, the memory layout changes necessary between bonded and non-bonded force calculations, because of the Fortran implementation, is not stable enough over long simulations. We should stop postponing this and transition finally to a Cython or pure C implementation.

Are we sure this is due to the Fortran implementation and not to the recently discovered OpenMPI memory leak issues (#186)?