If a vote (on a block or proof) has been going for too many rounds, increase the chances of the network coming to a vote by voting to reject after some threshold number of rounds.
The Avalanche whitepaper suggests that it's expected for finalization to occur after O(log(n))
rounds of voting. While our implementation differs in some ways, this expectation should hold
reasonably close. The threshold was selected to be an order of magnitude higher than the number
of votes required for finalization, which gives plenty of room for confidence to reset multiple
times before finalizing under normal network conditions.
This threshold can later be made configurable and/or tightened as needed. For now, this solves
the immediate issue of inconclusive votes hanging around for too long.