Use arbitrary precision multiplication for all signed multiplications in the interpreter because
it isn't easy to detect overflow. The previous test didn't actually work due to an optimisation in GCC.