TrainingContainer should keep track parameters used to generate stored quantities

[pmrv]

For FAIR data provenance it would be nice if exported TrainingContainers can know (roughly) how they were generated. This would include both generic dft parameters (k sampling, convergence criteria, etc.) and code specific inputs (ISYM, SYMPREC, ALGO, etc.). Ideally these could be taken from jobs directly once we have some generic setup inplace. This will probably take a while and TrainingContainer would need to store them in an efficient, sparsified way anyway. Therefore I will start here with a simple implementation first. Maybe just something like

DftInput= namedtuple('VaspInput', ['kpoints', 'electronic_convergence', 'code_specific'])

class TrainingContainer:
  ...
  def add_provenance(self, input):
    self._input_provenance = input

Then in a few steps it could be:

1. make it possible to do per structure in a sparse list
2. make it extract that automatically on include_job
3. export it into the pandas dataframe somehow as well
4. add optional checks that input parameters in one container are matching/compatible/consistent