First, an observation: the geometry in the paper is still slightly different. Your script doesn't show any geometry optimization although the N atom is placed slightly differently compared to the edge C atoms so maybe you did that before preparing the band structure calculation script? But even if so, your C-N bond length is a bit larger than in the article, 1.40 Å compared to 1.38. I doubt that this would change the whole structure from semiconducting to metallic though...
Now, I did run the calculation and got similar results to you for 9 k-points; in general I would not expect that you need as many as 200 points, and this was indeed the case.
I then had an idea: I changed to fewer k-points, in fact just 1 in the C direction. In this case I get the structure to be metallic! But is the result correct? I guess not, as it's just a result of bad k-point sampling. I can't be sure that this is THE difference compared to the paper, but one must always keep in mind the possibility that the reference calculation could be wrong - and upon careful reading of the paper it appears they use 3 k-points. I don't have time to rerun it now for 3 k-points, but you could test that yourself. And then, you may have a nice result to publish, with your 200 k-point result which contradicts the paper
