A Tor exit node can block destinations openly by declaring limitations in its exit policy (in principle by IP address or port, although I don't know if any exit policies use IP address this way) or by simply firewalling connections from itself to particular destinations. In the latter case, the node may be marked as a BadExit if a scanner detects that it intentionally blocked a connection that it offered to relay.

A node in an earlier position in a circuit can't know the ultimate destination of the circuit, or the content exchanged over it, although it could in theory blacklist individual nodes as a next hop (which doesn't typically really accomplish anything for content-blocking purposes, and could also be detected if it were a pervasive practice).

I think in HSv2 (the old hidden services protocol) there were a number of different opportunities for relays to selectively disrupt hidden service connections, while in HSv3 (the new revision) there are fewer.

A difference between Tor and IPFS in regard to what you're mentioning above is that in Tor, a specific entity (a specific clearnet or hidden service site operator) has to continue actively hosting a resource in order for it to remain accessible at its original location, while in IPFS, it's more like that there need to be some participants who haven't actively declined to host it, but they don't have to be the same as the original publisher.