"You can build arbitrary-precision rationals" is a far cry from saying, "this is a language designed for precision arithmetic." You can technically build arbitrary-precision rationals without BigInts using arrays, but it's not exactly ergonomic.
Even if we stipulate that users will be cognizant of floating point limitations and convert everything to integers, JS still requires all literals to be suffixed with n to avoid silently losing precision when integral values exceed 2*52.
