This is what I get when I set the observer height to 20m, and increase the "max distance" to 300km (200km = ~124 miles so may not be enough).
https://img.incoherency.co.uk/6478
It's also possible that the half dome is too short and the sampling rate of the line-of-sight jumps over it!
Heh, I almost hit back at the "in Rust" mention.
Would the end result have been different if it were done in python calling C libraries for performance? I strongly doubt it.