Enhancement: GROMACS FAHCore need to keep two levels of checkpoints.

[bb30994]

When a checkpoint is written to disk, there will often be a previous checkpoint. If it is renamed to checkpoint-old before writing checkpoint-new (and deleting 3rd checkpoint if an even older checkpoint if it exists) this will provide an additional level of redundancy. The various issues mentioning the guru meditation issues (which also need to be fixed) would become less critical if the restarting FAHCore could, upon recognizing that it cannot start from the most recent checkpoint could revert to the previous one.

This would also be essential in the cases where a cloud-based VM is abruptly terminated when it is preempted by a higher priority invocation.

[bb30994]

See Running FAHClient on a cloud resources on temporary VMs
https://foldingforum.org/viewtopic.php?p=333009#p333009

[shorttack]

I rate this as an enhancement to fix a design defect. @bb30994 is spot on about need for fall-back checkpoint.

[PantherX]

Any reason this can't be expanded to include OpenCL too? It would be wicked to have the consistent pattern in FahCore_22 as you can run GPUs in cloud compute and while not as common as CPUs, we're getting there so makes sense to future proof it now and have consistent behaviors across all FahCores.