MALA for multiple elements

[RandomDefaultUser]

While some work in that respect has already been performed at the Sandia National Labs, MALA does not have full multi-element capabilities yet. This PR aims to make multi-element systems fully available within MALA.
This includes:

• Two elements (example: Barium Oxide)
  • Bispectrum descriptor calculation (serial)
  • Bispectrum descriptor calculation (parallel/GPU)
  • Total energy evaluation
  • Total energy evaluation (using the atomic density formula)
  • Testing the full workflow (only that it runs, not that the predictions are good)
  • Implement a way to adjust the weights through the regular MALA interface
    • Make a plan to enable descriptor calculation on the fly for (bispectrum) descriptor optimization, because that could become pretty relevant
  • Confirming that the models are also accurate
• More then two elements (example: Barium Nitrate)
  • Bispectrum descriptor calculation (serial)
  • Bispectrum descriptor calculation (parallel/GPU)
  • Total energy evaluation
  • Total energy evaluation (using the atomic density formula)
  • Testing the full workflow (only that it runs, not that the predictions are good)
  • Confirming that the models are also accurate

While not the same issue in and of itself, it would be prudent to also tackle core electrons/"isolated" electronic levels in the same PR. I.e., if one has a system with, e.g., electronic levels located at -14 to -12 eV (core electrons) and then the next level is at, e.g., -1 to 0 eV, than by default the LDOS contains a lot of zeros. This can be alleviated by

• Implement an LDOS cutting that cuts out zeros in the middle of the LDOS
• Test that the entire worklfow runs with that

[RandomDefaultUser]

I just realized that the TEM will work by default, since there is only one pseudopotential path anyway. The atomic density/Gaussian equation of course has to still be tackled.

[RandomDefaultUser]

I confirmed basic functionality for the two element case. The code changes so far have been minimal, but the accuracy is not yet great. That could just be because I am using a minimal example (https://github.com/mala-project/test-data/tree/multielement, Barium Oxide unit cell), or because the bispectrum descriptors need tweaking. I will investigate this further and look into the three element case.