I've switched over to using primary/replica for database stuff because it's more accurate. The replicas don't always behave themselves so calling them "slaves" implies a level of obedience to the "master" that they don't have.
For a distributed database there is also fragmentation/sharing though. In this case calling the nodes replicas is not accurate. I guess you would call these “shard” or “dsta” nodes.
You are right about the “slaves” not behaving, in fact they jump on the chance to become the “master” themselves once the current “master” goes down. Then there is the split-brain problem.
It's really more like a worker boss relationship, but I would hesitate to call database nodes “workers” because this one is usually used for a processing engine like Spark.
In Cassandra all nodes are equal, but specific nodes hold specific parts of data. The only concept is the coordinator node which is the node that got the query. All nodes can handle all queries. The coordinator node is different per request.
Usually quorum is maintained by using odd numbering so that when one node fails the next sequential node assumes the role allowing quorum-election breaking any ties.