[krbdev.mit.edu #1353] Decide on a long-term krb524 strategy

[Sam Hartman via RT]

Do we want to move this into libkrb5?  One disadvantage would be that
our ABI would be different when built --without-krb4 than when boult
--with-krb4.


If not, do we want to install a header and make it a shared library?


Also, we need to deal in some manner with the fact that the error
table initializer is needed on Unix but not exported on KFM.