Non-reproducible behaviour of `addMissingAtoms`

[JoaoRodrigues]

Running addMissingAtoms() gives different results because the LocalEnergyMinimizer does not allow setting a random seed from pdbfixer. The end result is that re-built sidechains are pointing in different directions at every different try, which is not good for reproducibilty. Setting random.seed and np.random.seed does not help. The following code snippet ran on any structure missing heavy atoms will reproduce the problem.

for i in range(5):
    random.seed(917)

    import pdbfixer as pf
    from simtk.openmm.app import PDBFile

    mol = pf.PDBFixer('../example/4xt1_AB.pdb')
    mol.findMissingResidues()
    mol.findMissingAtoms()
    mol.addMissingAtoms()

    with open('output_{}.pdb'.format(i), 'w') as handle:
        PDBFile.writeFile(mol.topology, mol.positions, handle)

My suggestion would be add an optional seed parameter to addMissingAtoms that then sets the integrator seed. I'll follow with a pull request.