Fix band stats k-point labels

[utf]

Summary

This PR fixes a long-standing (since 2022) bug in sumo-bandstats in that the labels for the effective masses don't show correctly. E.g., this is the output for the band stats in test/data/Cs2SnI6/bandstructure:

  m_h: -0.641 | band 35 | [0.00, 0.00, 0.00] (\Gamma) -> [0.00, 0.00, 0.00] (\Gamma)
  m_h: -0.516 | band 35 | [0.00, 0.00, 0.00] (\Gamma) -> [0.00, 0.00, 0.00] (\Gamma)
  m_h: -0.641 | band 36 | [0.00, 0.00, 0.00] (\Gamma) -> [0.00, 0.00, 0.00] (\Gamma)
  m_h: -0.516 | band 36 | [0.00, 0.00, 0.00] (\Gamma) -> [0.00, 0.00, 0.00] (\Gamma)
  m_h: -1.234 | band 37 | [0.00, 0.00, 0.00] (\Gamma) -> [0.00, 0.00, 0.00] (\Gamma)
  m_h: -15.158 | band 37 | [0.00, 0.00, 0.00] (\Gamma) -> [0.00, 0.00, 0.00] (\Gamma)

Obviously the effective masses are not in the Gamma -> Gamma direction (which doesn't make sense). The output for the new version is:

  m_h: -0.641 | band 35 | [0.00, 0.00, 0.00] (\Gamma) -> [0.50, 0.50, 0.50] (L)
  m_h: -0.516 | band 35 | [0.00, 0.00, 0.00] (\Gamma) -> [0.50, 0.00, 0.50] (X)
  m_h: -0.641 | band 36 | [0.00, 0.00, 0.00] (\Gamma) -> [0.50, 0.50, 0.50] (L)
  m_h: -0.516 | band 36 | [0.00, 0.00, 0.00] (\Gamma) -> [0.50, 0.00, 0.50] (X)
  m_h: -1.234 | band 37 | [0.00, 0.00, 0.00] (\Gamma) -> [0.50, 0.50, 0.50] (L)
  m_h: -15.158 | band 37 | [0.00, 0.00, 0.00] (\Gamma) -> [0.50, 0.00, 0.50] (X)

Details

This bug was introduced when trying to fix another bug (#169), specifically that the k-point indices listed in sumo-bandstats were wrong.

This all comes from the fact that the pymatgen logic for extracting branches in the band structure requires duplicating the high-symmetry points. To avoid wasting compute resources on these extra points, sumo-bandplot/bandstats adds this points dynamically rather than including them in the DFT calculation. However, this means sumo-bandstats reports the wrong indices for the VBM and CBM with respect to the band structure calculation. The previous PR tried to fix this issue by removing the branches for sumo-bandstats but this then led to the labels being missing.

The fix in this PR is to:

1. Revert the change in Fix k-point indicies in sumo-bandstats #169 and instead always keep the branches (to get the correct labels).
2. Calculate a mapping for the k-point indices from the branched band structure to the original band structure indices. Use this mapping when logging the band stats information.

Other fixes

I also added a quick fix the VASP bug where it sometimes puts the Fermi level in the valence or conduction bands when using ISMEAR = -5. I already added a workaround for this in pymatgen (specifically vr.get_band_structure(line_mode=True, efermi="smart")) but it currently isn't used in sumo.