I'm just talking around all the theoretical stuff of protocol design here. How I would do it:

* shove the defmt data into a byte stream, not cobs encoded
* have one piece of your code consume outgoing space packet data and outgoing defmt data
* when you want to send defmt data, just grab a chunk of defmt data, as much as is available, or the max frame size, and stick it in a packet, encode it the same way you encode every other frame