Store optimization steps in large parallel runs

[WardLT]

Our current implementation for storing the energies and forces at each ionization step has each process in the workflow running connecting to a single file. Connections from many processes leads to problems with the locking system ASE uses for their database.

A few options:

• Switch from a file-based system to storing data in a service (i.e., something designed for concurrent access)
• Only write to the database from a single projects (e.g., send data back to the thinker before writing)
• Don't write to a global store, instead write the DB in the run directory (which should be saved Save all output files #65)