[Py] Real twist produces complex orbitals

[bkesk]

This is migrated from an issue by Ryan Levy. I updated the code snippet based on changes in afqmctools. I'm not 100% convinced that this is an issue - there isn't really a guarantee that the orbitals should be real even if a no twist is used. Keeping it for now anyway.

Due to imaginary noise in the Hamiltonian with a real twist, the returned orbitals are accidentally complex
MWE:

import afqmctools.systems.lattice as lat
import afqmctools.hamiltonian.model.builder as ham
import afqmctools.utils.io as io
from afqmctools.wavefunction.free_electron import free_electron

import numpy as np

lattice = lat.SquareLattice(
            L=(4,4),
            axis1_boundary=lat.PBCBoundary,
            axis2_boundary=lat.PBCBoundary,
            twist=["1 Pi","1 Pi"]

)

params = {
    "hamiltonian": {
        "U" : 2.0,
        "t" : 1.0
    },
    "lattice" : {
        "L1" : 4,
        "L2" : 4,
        "boundary1" : "pbc",
        "boundary2" : "pbc",
        "twist" : ["1 Pi", "1 Pi"]
    }
}

wfn, _  = free_electron(params, (8,8), lattice=lattice)
print(np.max(np.abs(wfn[1].imag)))

produces:

0.2500000003735068