I think it may depend on how well distributed would a service be: having several big servers would not help but if every family and company had their own mini server, located in a non-censoring country then the censors would be unable to do anything easily. These servers, in turn, would be able to easily connect to the broader network. Of course that wouldn't be as easy to setup as a simple installation of the Signal app.
An aspiring censor could also "easily connect to the broader network" and masquerade as a federated server in order to discover others. This process could even be automated.
Federated services also require an identifier, and this identifier usually indicates where the user's account is located and how to connect with them (e.g. user@domain.com). As people share these identifiers, the aspiring censor can just keep adding new entries to the blacklist.
At least in case of XMPP, the client doesn't need to be able to connect to other domains, so as long as you can connect to your own server outside of the censorship's reach (which could be accessible for c2s connections in a completely different way than for s2s), you should be fine.
Yes, but it's always possible to block IP (targeted attack). Federated with a big amount of small servers make it hard to automate. You can block several hosts but the rest of the network would work fine. And because of how federation in XMPP works you just need a one client to server connection to reach the entire network.
