Ideas for phonon workflow

[JaGeo]

• Move the structure optimization to the very beginning of the flow
• Make the pre-optimization more robust (custodian?)
• Have a more robust decision on what volumes to benchmark the phonon computations on for qha
• Check supercell sizes in advance of the submission of the flow and warn users
• Make data reuse easier (i.e., automatically export all benchmark structures into a json or similar at the end of a flow)
• Simplify the benchmark step and start from a list of structures and ML models