For fun I created a sample of component symbols based on your data:
The height is approx 50 mil, some western text shown for comparison. It seems fairly readable, I would have to leave it to language experts to properly judge that. I wasn’t sure of the scaling of the data so that may have messed up the rendering.
I think that scalability would have to be considered, whether the data is in text or binary. Text makes an easier way to compile in of course. I think I would at least split the Hershey data into manageable blocks. The Hershey data could be packed more, but the source files might get bigger. Using null-terminated strings has some advantages, so if you could avoid use NULL character there are 255 values to play with.
CJK_font.lib (86.1 KB)