feat: implemented BlockHeader (BFT-357)

[pompon0]

Implemented https://www.notion.so/matterlabs/5d238e08fbed476b8550f697d9c39df6?v=51a4bffaba05499e8ca11bd625fff54a&p=16c93154366c4a3994457b44043dea71&pm=s with some differences. Summary of changes:

• Replaced Block with BlockHeader with just a hash of the payload
• Made consensus messages include the whole BlockHeader, rather than just hash (since it is small, once we moved the payload out)
• Introduced a strongly typed Payload
• Added protocol_version to the BlockHeader (and relevant checks to the consensus) - currently hardcoded to 0

Minor fixes:

• split the leader::Error and replica::Error into per-method errors (since the error types returned by each are disjoint)
• relaxed the "stringly" typed assertions from sync_block tests (if needed they should use strongly typed errors instead)
• added "size" to the protobuf encoding of BitVec, so that the encoded BitVec value is context free

[brunoffranca]

Looks good, I only have a few minor nits.

[brunoffranca]

Instead of letting it as a TODO, let's think about it now. What are the arguments in favor? (I assume the only argument against is bigger messages)

[pompon0]

The argument in favor is that if we have a new version which changes the semantics of the consensus message it may happen that the validator running old binary will sign a message that will be incorrectly interpreted by a validator running a new binary.

[brunoffranca]

I think we should do it then.

[brunoffranca]

We discussed this on the last team meeting and were in agreement that we should keep just one version for both consensus and execution layers. I said that I was going to try to find some reason to have separate versions, but I couldn't find any.
Do you still feel that we need to think this through more or are you happy committing to one protocol version?

[pompon0]

I had some afterthoughts after that discussion in the context of the above: if we keep the version (only) in the block, then some messages don't include the version at all. And if we (hypothetically) added version to both header and the messages, then forcing them to be the same wouldn't bring any benefit and might be constraining. Not a strong opinion, just writing down my thoughts.

[brunoffranca]

Well, if we include the version in every message, then we don't need to include it in the block header. The block header is never sent by itself over the wire, either it is sent in a leader prepare or in a final block (which would contain the version in the commit QC). It would even make some sense since the protocol version isn't technically a property of the block, but of the nodes.