Improved checks to sequential parameters in OpenMM simulation schedules

[timbernat]

Have documented a handful of issues with loading of serialized States and systems in the sequential openmm simulation runner that need to be resolved:

• Ensemble-specific system force parameters (namely Barostats) are propogated where they shouldn;t be (e.g. a schedule which specifies NPT -> NVT will actually end up being run as
• Currently no check for when prior state is empty / invalid upon read
• No protocol for interruption if a simulation fails along the schedule
• Integrators are currently not being serialized to XML as part of openmmtools.serialization, nor are they supported by SimulationPaths; plan to incorporate these into the workflow templates