ReplicationCoordinatorImpl::fillIsMasterForReplSet should return isMaster:false while in shutdown

Clients use the isMaster field here to determine whether or not to direct writes to the node. When we're technically "master" but we can't accept writes (i.e. when in drain mode), we claim not to be master. We should do the same while shutting down.

This affects retryable writes, in that a write interrupted due to shutdown may immediately re-try on the same node if it still claims to be master.