I’ve read that the Desmume core uses a software renderer, but could conceivably be ported to the hardware renderer.
Do I understand the situation correctly? If so, how accessible might this task be for a new contributor?
Depends on whether you’re an experienced programmer who just hasn’t contributed to libretro before or are also a new programmer.
The task itself isn’t that terrible, theoretically, but you’ll need to have an understanding of some aspects of OpenGL and passing the core’s framebuffer off to libretro for the frontend to display.
I abandoned this idea because melonDS began advancing rapidly and it seems like the libretro core is staying pretty close to upstream.
By the time I finished this, a more promising core would likely be available soon after.
I don’t think we have the 3D renderer hooked up with the libretro core currently, so if you would like to work on that, we would welcome it 
You mean melonDS? Interesting!
Start a bounty for it (Desmume GL rendering) and see if there is an interested developer who wants to do this.
I doubt melonDS will be at Desmume’s level for at least another year and a half.