Light "diode" and 2nd law of thermodynamics
Light diodes exist, and yet the 2nd law endures. In general, any perfect ratchet would allow you to violate the 2nd law. Since very few Nobel prizes have been awarded for perpetual motion machines, experimental evidence seems to suggest there is no such thing as a perfect ratchet.
When confronted with any type of device that rectifies the flow of energy, the appropriate question, in my opinion, is 'how is this device imperfect'? In the case of a Faraday isolator, Lord Rayleigh addressed this question a long time ago.
EDIT:
The other reference at the end of the Wikipedia article on Faraday isolators is a pretty nice discussion of this topic.
A light diode that would reflect every photon coming from one side but allowed all photons from the opposite side to be fully transmitted - and that wouldn't absorb anything - would be in contradiction with the second law because that would be nothing else than a photon version of Maxwell's demon: it would be able to collect all particles (photons in this case) in one half of a vessel.
Such things can't exist because the second law demonstrably holds for all objects with many degrees of freedom.
None of the "almost light diodes" from the real world that you mentioned is doing the same thing - mostly because the objects you mention absorb a sufficient fraction of the incoming light from one of the two sides. When they absorb the light, they turn the absorbed photon's energy to heat - a more chaotic form of energy - and increase the total entropy by a sufficient amount so that the overall entropy trend is never negative.
The simplest "light diode" (optical isolator) (link) is two polarizing sheets 45 degrees apart, with a 45 degree Faraday rotator in between. The polarizing sheets absorb light with one polarization and transmit it with the opposite polarization.
To check the second law of thermodynamics, you need to assume the polarizing sheets are at the same temperature as the other things you're thinking about. Therefore the polarizing sheets will emit their own blackbody radiation, always polarized in the direction that they absorb light. The device violates the second law of thermodynamics ONLY if you FORGET the blackbody radiation of the polarizing sheets.
Things get more complicated when you replace "polarizing sheets" (which emit blackbody radiation) with "polarizing beamsplitters" (which don't). A polarizing beamsplitter sends each light-polarization in a different direction, for example it might back-reflect one polarization and transmit the other. But no matter where you send the non-transmitted light polarization, you will always end up causing problems for your perpetual motion machine. You can try drawing it out to see the various ways this happens. This paper (linked by Andrew above) goes through some of the possibilities in detail. :-)