Implement an asynchronous BFT algorithm

[kayabaNerve]

HoneyBadgerBFT (https://eprint.iacr.org/2016/199.pdf) is a BFT protocol which notably can be run over Tor. If we ever want to hide the IP addresses of multisig participants, we would need to move Tributaries over Tor which Tendermint would not be viable for. An asynchronous BFT algorithm, such as HoneyBadgerBFT, would be viable.

Since HoneyBadgerBFT, the field has been improved with:

• Aleph https://arxiv.org/pdf/1908.05156.pdf
• Dumbo https://eprint.iacr.org/2020/841
• Speeding Dumbo https://eprint.iacr.org/2022/027
• Dory https://eprint.iacr.org/2022/1709.pdf

I cannot comment on which is preferable at this time.

HoneyBadgerBFT specifically requires a trusted setup or synchronously performed DKG. Accordingly, we'd have to perform a DKG over Serai itself (preferably with off-chain blobs) to set up each Tributary.

Tributary itself would also have to be rewritten over a TX stream.

Supersedes #276, complicates #286, see #271 for a discussion of some relevance.

[kayabaNerve]

Narwhal/Tusk would also be a candidate: https://arxiv.org/abs/2105.11827

[kayabaNerve]

https://eprint.iacr.org/2024/142 has the lowest round complexity thus-far.

[kayabaNerve]

https://eprint.iacr.org/2024/653 is a new work on the matter.

[kayabaNerve]

https://eprint.iacr.org/2024/1235 on the liveness of Tusk and a necessary modification to prevent their posited attack.

[kayabaNerve]

My criteria for which we move forward with will probably whichever is simplest to understand and audit so long as it doesn't have unbearable performance. Tributaries are fine with latency on the scale of a few minutes so long as we have sufficient throughput.