Add an equilibration phase for slow-equilibrating properties

[lilyminium]

Related to #591. As discussed in the FF fitting meeting and on Slack, one solution to slow-equilibrating properties would be have an "iteration 0" in FF optimization and benchmarking that allows the use of previously simulated boxes to provide initial coordinates.

This solution should ideally:

• generalize broadly and robustly to unseen properties (i.e. there should be no danger of similar but different properties contaminating a run)
• generalize across force fields
• allow use of previously-computed data outside of the current Evaluator run
• work across optimization and benchmarking