I think it's useful to talk about a real concrete example, so thank you for this comment.

Ideally, people wouldn't send death threats/hate speech. This seems practically unenforceable from Twitter's side.

So what if Twitter had a special feature where it could detect if you are receiving an unusual spike in notifications (and possibly even use sentiment analysis). Then it would present you with a nice, warm message:

"We noticed you're receiving an unusually high number of notifications. These messages may be abusive, hateful, or triggering."

Then provide options for turning your profile private, or a special new setting that completely hides @-notifications from people you don't follow.