Have you never seen under the hood of an NES (or an SNES for that matter)? The system can only display a small number of sprites (and further limit of 8 per row), so most of the graphics are drawn via the map, which is a grid of 8x8 tiles that all move together. In theory you could certainly load a full-screen bitmap that's been cut into pieces 8x8 big, but that would use up most of your cartridge (besides being more complicated to set up) so generally, the same tile is used in more than one place on the grid, hence all the UI elements being a multiple of 8x8 (or at least a multiple of 8x8 apart from one another).

Interesting challenge, writing your own parser...! I was tempted to say that going out of your way to make something more like ZZT when it doesn't have to be is the definition of masochism, but... stripped of all the legendary workarounds that it usually requires, the language itself seems fine! (Though ":END #END :DIE #SHOW DEAD" does seem like the sort of program Murderbeaks would write....)