That's not a problem for voice calls because voice calls inherently require both participants to be online.
Though I am curious why Signal's approach [0] wouldn't work for Threema.
Using plain NaCl boxes also has certain advantages: The crypto is quite simple and you can encrypt a message statelessly for a recipient if you have their public key.
With the Signal protocol, if I understand it correctly, you need to pre-generate and exchange a number of keys. With this process, I think you can run out of keys if you encrypt a lot of messages without the other party being online. And you need to exchange those keys before you can even communicate with each other.
After the first back and forth communication, it's encrypted forward securely, with a full rekeying happening in each back and forth interaction. This does involve state, yes, but it works 100% asynchronously. In fact I think it's also compatible with the multi device plans that Threema has.