On a disaggregated-storage standby, two unrelated threads can both decide to start oplog
application at nearly the same instant. The second one calls
OplogApplicationCoordinator::start() on a coordinator that is already kRunning and
trips invariant(_state == State::kDrained), aborting the node (SIGABRT). The fix makes
start() idempotent: a kRunning coordinator is a no-op instead of a crash.