[krbdev.mit.edu #7604] Dynamically expand timeout when TCP connects

[Greg Hudson via RT]

I don't think this implementation takes effect when the TCP connect 
returns 0 immediately (as opposed to EINPROGRESS).  Probably not 
important, but I'm noting it here for the record.  If we switch to a 
rule-based implementation of sendto_kdc as suggested in 
http://k5wiki.kerberos.org/wiki/Projects/Rule_based_sendto_kdc_loop this 
would be pretty easy to fix.