On cassandra-cli if you "describe cluster;" I guess you will see an "UNREACHABLE" node.
If you do so, there is a way to remove this unreachable node.
Go to the JMX management console (ip_of_one_up_node:8081 by default)
Then go to the org.apache.cassandra.net:type=Gossiper link and use the "unsafeAssassinateEndpoint" input. Fill it with the ip of the down node and invoke the function.
"nodetool gossipinfo" should now tell you that this node has left the ring and let you truncate or whatever you need to do.
Use this carefully this function is composed of the "unsafe" and "assassinate" words because it forces the node to go out of the ring without any check or replication.