Why does FK4 <-> Galactic disagree between Astropy/Kapteyn and PyAST/PySLALIB

@timj and @dsberry: Astropy and Kapteyn currently agree well with PyAST/PySLALIB for the FK5 <-> Galactic conversion, with differences around 1mas. On the other hand, for FK4 <-> Galactic the differences can be up to 0.2 arcseconds which is quite large. This is a bit weird since the FK4 <-> FK5 conversion agrees well between all packages:

http://www.astropy.org/coordinates-benchmark/summary_matrix.html#astropy

I think that both Astropy and Kapteyn do:

FK4 -> FK4 (no e terms) -> Galactic

Do PyAST and PySLALIB do something different? What would cause such a large difference? Why do we agree for FK5 -> Galactic but not FK4 -> Galactic? (the relevant scripts are in e.g. `tools/kapteyn/convert.py`, `tools/pyast/convert.py`, etc.