Fix trajectory indexing issue on restart

[lohedges]

This PR fixes a strange bug where trajectory indexing fails when attempting to save trajectory frames at a checkpoint after restarting a simulation. It appears that the issue must be related to some stale trajectory object being associated with the stream file, since calling mols.delete_all_frames() prior to the simulation fixes the error. I'm not sure why this happens, since we do delete the frames after checkpoint during the simulation and this issue doesn't occur in all cases. I've added in some safeguards to avoid this issue, i.e. we now do the following:

• Delete frames from the copy of the system that is streamed to file.
• Delete frames in the system used for dynamics after checkpointing.
• Delete any existing frames in the input system prior to simulation.

closes OpenBioSim/cresset#80