kadmind patch (was: kadm5_randkey_principal_3)

[Sam Hartman]

Thanks for identifying the problem.  This patch certainly would fix
things, but I can't help wonder whether it wouldn't be better to adopt
a solution that involved less code duplication.

[more directed internally] Would it be reasonable to have all
self-changes go through the _3 function, passing in a default ks tuple
set to the wrapper in cases where the called RPC didn't support that
parameter?