Retry SASL writes if buffer not fully sent
    https://bugzilla.redhat.com/show_bug.cgi?id=518544
    Resolves: bug 518544
    Bug Description: large entries cause server SASL responses to fail
    Reviewed by: nhosoi (Thanks!)
    Branch: HEAD and 1.2
    Fix Description: The SASL server code was broken when we switched over to
    use NSPR I/O for the SASL IO layer.  If the entire encrypted buffer could
    not be sent to the client, the server was just failing.  Instead, the server
    must keep track of how many encrypted bytes were sent.  If all of the
    encrypted bytes could not be sent, we must return the appropriate error
    to the caller to let them know the operation would block.  The caller in
    this case is the write_function() which does a poll() to see if the socket
    is available for writing again, then will attempt the send again.
    I also cleaned up usage of the various Debug macros.
    Finally, I discovered that the sasl init code was calling config_get_localhost()
    before that value could be set.  In most cases, it is ok, because it will
    fall back to the default hostname from the system.  However, if for some
    reason you want to use a different localhost, it will fail.  Now it will be
    set in the boostrap config code.
    Platforms tested: RHEL5 x86_64
    Flag Day: no
    Doc impact: no