[krbdev.mit.edu #2266] wrap_size_limit broken for CFX

[Ken Raeburn via RT]

Well, I'm still not seeing the off-by-one issue, for some reason, but
this version of the code seems to get the right answer (including your
example of 1076 -> 1016), so I'll be checking it in.  Thanks for
catching this.

Ken

    while (sz > 0 && krb5_encrypt_size(sz, ctx->enc->enctype) + 16 > req_output_size)
        sz--;
    /* Allow for encrypted copy of header.  */
    if (sz > 16)
        sz -= 16;
    else
        sz = 0;
#ifdef CFX_EXERCISE
    /* Allow for EC padding.  In the MIT implementation, only
       added while testing.  */
    if (sz > 65535)
        sz -= 65535;
    else
	sz = 0;
#endif