This is an extension of
ISPN-1896 of sorts, but for all the cache operations that are visible to the user.
After a node leaves, the other nodes that have sent commands to that node should either ignore SuspectExceptions or, if not possible, they should retry the operation (e.g. if they didn't get any response back).
For example, VersionReplStateTransferTest quite often on my machine with a SuspectException, because the versioned prepare command expects a response from the coordinator and the coordinator has just left.