I think the idea is that attendants are much easier to train and thus replace quickly than drivers are. All they have to do normally is push a button saying "it's safe", more or less. Perhaps very rarely they'd need to go into the tunnels to reach a stuck train, but that's exceptionally rare. In a squeeze people could be trained to do the normal tasks in a day or less.

DLR safety - aren't you agreeing with me? The DLR is not perfectly safe but in reality people accept it. Nobody is clamouring in London for the DLR to be shut down on the grounds that it's a death trap. London society has made its peace with the level of safety the DLR provides. Would more safety be nice, of course. Is it actually a hard requirement that blocks other upgrade projects? That's not at all clear and the argument you're making here is the same circular one the article makes - automation upgrades require safety upgrades because the regulator (+other undefined people) say they require safety upgrades, and the regulator says it requires safety upgrades because automation requires safety upgrades. But the link between these things seems rather brittle.