Pluggable configuration design choices

[Sam Hartman]

A couple of comments.

First, I'm nervous about having additional environment variables that
take paths to dynamically loaded modules.  I think this is not a huge
concern because you can already accomplish this through krb5.conf,but
this causes me to prefer requiring config modules to live in a special
directory.  On the gripping hand, that probably makes regression tests
in -tree difficult.

Second, are iterators required for capaths or does that just require
getting all values of a given key?