By default when you install the agent on any single node the installation recognizes that it is a cluster and presents you with a list of available nodes from which to select.
With a RAC cluster you have one database and multiple instances. All instances will write to one, and only one AUD$ and FGA_LOG$ table so if you are using database auditing one node will suffice. Which node that is though depends on knowing which node is up at any one time so you could potentially choose a node that is dropped from the cluster or is down for patching and maintenance while the cluster is still running. With respect to REDO collection each node has its own redo thread so you definitely need to be collecting from every node.
Audit Vault has not been out long enough that I can tell you from experience what might be defined as "best practice" and often what we really need to study is "worst practice" to know what not to do. But in the case of RAC my instinct would be to first determine the collection type(s) and then decide. Erring on the side of collecting from all nodes makes a lot of sense.