This pathology is not peculiar to the potential step problem, or 1D problems even. It is a general feature of expanding solutions of Schrodinger equation in plane waves (or their equivalent for some other potential). These solutions are not normalisable, so technically they are not member of the Hilbert space of states. You are right. These states are not physical. But they can be linearly combined into wavepacket states which are normalisable.
There is a mathematical formalism making all of this rigorous, but for most physical purposes it is unnecessary since at the end of the day all you care about are wavepackets. Look up "rigged Hilbert space" for more information.