Implement recovery mechanism for "out-of-sync" nodes

[sbellem]

NOTE: This is in a way a duplicate (or a refinement) of #33 (originally amiller/HoneyBadgerBFT#6). Perhaps the two issues can be merged into one.

Using the block signature mechanism outlined in #15, nodes that need to catch up should be able to do so.

From amiller/HoneyBadgerBFT#57:

After finalizing each HBBFT-Block, nodes produce a t+1 threshold signature on a merkle tree over all the transactions, as well as a merkle tree over the current state file S. This signature serves as a CHECKPOINT, a succinct piece of evidence that the current block has concluded. CHECKPOINTs are used for three different purposes:

...

• II. Allows lagging (or restarted) nodes to recover. When a node receives a valid CHECKPOINT for a later block than the current one (for block B’ > B), then it determines it has fallen behind. It deallocates any buffered incoming or outgoing messages in block B, and

...

Proposed name for this mechanism: speedybadger