[krbdev.mit.edu #6284] memory leaks in error conditions

[Greg Hudson via RT]

First patch looks fine; I adapted it to the current code (which was
changed fairly heavily by Luke) and will commit shortly.

I don't understand the second patch.  The first hunk appears to be for
Apple-specific code, so I'm ignoring that.  The other hunks move a
free(reply.enc_part.ciphertext.data) into the errorout label.  But (a)
that move seems unnecessary, since in the old location the data was
freed immediately after it was allocated (i.e. on successful return from
krb5_encode_kdc_rep), and (b) that move seems incorrect, since "goto
errout" can happen in cases where reply.enc_part.ciphertext.data hasn't
been initialized.