We never do 160-bit arithmetic.
Introduce new opaque implementation of `uint256`, move old "arithmetic" implementation to `arith_uint256.