Hmmm... it worked
Possibly not the most practical but it was an interesting concept to try out.
A couple of things to note. Whilst I modeled the moon with relatively realistic numbers the default sunlight diameter of 0.5° was just a little wide (the moon should have been a tad wider than 0.5°). Dropped it back to 0.49°. While visible disc was not checked, the diameter still has an effect on where the glow starts. Setting it to something really small gives you a bright point of light so the best compromise IMO is to match it to the diameter of the moon.
The problem in my images above is that the shaded part of the moon reduces the density of the glow effect. Solution: Only render the portion of the moon lit by the sun.
1. Create square image with a white circle.
2. Load into PTGui and set as a 185° circular fisheye image (a few extra degrees to catch mountains/craters on the terminator when I add the bump map)
3. Set its yaw and pitch to match the heading and elevation of the sun
4. Create 360x180° equirectangular image
Add this image to internal network of moon, spherical projection, same position as moon. Connect to opacity function of default shader.
Fixes one problem but it would also allow stars behind the dark side of the moon to show through, although with a bit of luck the atmospheric glow will be high enough to disguise it. It also means you have to create a new mask image whenever the sun changes position (although there are ways to do this on the fly for animations)
For phases smaller than a half moon it may be necessary to switch to a rectangular source image (equirectangular projection) for PTGui to create a mask to render just a wedge of the moon, but there may be other factors for why this may not be a good idea in that situation.