That's a separate nvim instance within the terminal of the outer nvim instance. If I'm understanding this feature correctly, the idea is that instead of that, you make the outer nvim instance do stuff. Then pasting back and forth between buffers works better, etc. It wasn't clear to me from this short blurb though if that includes blocking for the buffer to be handled and then signalling success or failure to the caller. (or for that matter if I'll remember the mechanism for it if there is one. I operate on muscle memory and might just stick with the tiny editor within the bigger one. But I'm curious anyway.)
It could even overlay the terminal pane for the duration of the command with the buffer pane. So it'd look a lot like opening the second editor instance, except incorporated with the first editor's buffer list, registers, etc.