Fix pubsub broadcast loops

[Stebalien]

See https://github.com/libp2p/go-libp2p-pubsub/issues/573.

Tasks:

1. [ ] Write a spec change.
2. [ ] Implement it in go-libp2p-pubsub.
3. [ ] Integrate into F3.

[Stebalien]

I still really want to do this but... I'm not sure how critical it is now:

1. We won't re-broadcast messages for at least 2m (the time cache).
2. We expect an instance every 30s.
3. We drop messages from prior instances.

That means this is only an issue when we have a long instance. But in that case, we kind of want to rebroadcast and, if we're worried about exponentially increasing traffic, we should probably just drop messages from old rounds instead.

[Kubuxu]

It makes sense, given that we have to be able to handle the parings from N participants either way.

[Stebalien]

We can also fix this in a future network upgrade without too much difficulty.

[Stebalien]

We're deferring this because it's not a priority. Instead, we're focusing on #583.