Investigate improving cosign fault tolerance using random subset sampling

[kayabaNerve]

Instead of cosigning being based on validator sets, we should investigate if random subset sampling would enable us to maintain a fault tolerance of 33%. This would increase the computational cost of verifying cosigns, yet would notably reduce the cost of producing cosigns.

We could also restore fault tolerance by removing the validator-set-based cosign, yet then we have cosign performance linearly decrease with horizontal network growth :/

[kayabaNerve]

If we haven't observed a cosign in however long, we can fall back to a non-horizontally-scalable live cosigning protocol.

So we attempt the horizontally scaling, 12% DoSable protocol, yet on omission, we fall back to each validator posting a cosign (waiting for 67%, 34% DoSable).

I do like the idea of introducing this pessimistic path. Credit to Jeff Burdges for the suggestion.