count equivocations on all votes

[rphmeier]

Closes #34

The previous behavior was that equivocations would be counted on all branches that were actually equivocated for. Now, if we see a voter double-voting, we count it as "infinite-voting" and treat it as if we have received a vote for every possible block. This means that we can ignore any equivocations beyond the first.

Rewrites most of the bitfield module. The idea is that now, each VoteWeight is its own bitfield which counts all the voters who have voted on a specific node in the graph, and there is also a round-global bitfield of all equivocators.

When calculating the vote-weights corresponding to a node, we calculate the union of node_bitfield and equivocators_bitfield, and figure out the total weight based on that. I couldn't figure out a more allocation-efficient way of doing this. A pool for bitfield storage might be an improvement.

Since we now have a need for a global-ish lookup of validator indices to weights (before it was only equivocators -> weights), I introduced the VoterSet type which is used in lieu of HashMap<Id, u64> in the codebase.

[andresilva]

lgtm, it's simpler since there's no need to special case the double-counting. currently we may produce invalid commit messages since on equivocations we only include one of the equivocated votes that justifies the commit, whereas now equivocated votes justify any commit for that round and should always be included. I guess this can be addressed with #37..