Persist current hbbft state on regular shutdown

[dforsten]

A regular shutdown like:

• Rebooting the server
• Stopping/restarting the client for the purpose of a parity client upgrade

should not cause the node to become faulty for the current hbbft round/epoch.

This could very quickly push the chain into a state where more than f nodes are considered faulty (and thus stalling the chain), on operations that should be considered normal in real-world production environments.

To achieve the required level of robustness we will have to persist vital hbbft state information to disk and load it on the next validator startup.

This includes at least:

• The current hbbft state
• The hbbft consensus messages currently cached in the sync module, but not sent yet

[dforsten]

As discussed in poanetwork/hbbft#421 we need to persist both outgoing and input messages (along with the rng seed) for active epochs to have all necessary data to deterministically replay the epoch to restore the hbbft state as it existed before client shutdown.