The collective body that's deciding which version of code to run has a fiduciary responsibility. To some extent, that includes everyone, but in reality, it's a relatively small group of people.

You can't know with certainty what will happen without trying it, but you can gain an increased level of confidence by doing formal analysis of the change, and developing research-backed theories that are more reliable than our intuitions. You can also increase your level of confidence by running simulations.

Flag wars, much like self voting cannot be countered through code changes

Where is the evidence for this? If self-voting can't be countered through code changes, there's no point in implementing the change. As suggested in item (iii), however, I suspect that they actually can be mitigated by realigning the voting incentives.