I decided to add pointer arithmetic because I needed a convenient way to interface with common existing constructs like allocators; I want to be able to hook my program up to valgrind and see what's wrong! But these pointer arithmetic operations are also generic across implementations: the web implementation uses tape indices as pointers (i.e. pointers with element sizes of 1), but the desktop implementation uses malloc/free and uses regular eight byte pointers. The compiler doesn't know the difference! This allows the architecture a large amount of flexibility across lots of different types of backend implementation!
Thanks for the feedback!