Fix issue #84

[lohedges]

This PR closes #84 by fixing the restart handling for GCMC simulations. The specific issues that have been resolved:

• System comparison checks on restart now account for the fact that the checkpoint files contain additional GCMC waters.
• The gcmc_checkpoint frequency is allowed to change between configs when it was originally None, since it will be updated to the energy_frequency in this case.
• When re-validating the gcmc_selection for a restart, we need to account for the fact that self._system will be a list of systems.
• I had incorrectly deleted all ghost waters from the restart system, rather than only the additional GCMC waters that are padded on the end. We need to do this so that we can set up the GCMC sampler in the normal way, following which the GCMC state of the restart is applied to the system, i.e water state and positions.
• The final checkpoint for the regular Runner is now consistent with the others, i.e. it performs a backup and uses the lock file.

@tom-potter-cresset