Add the ORConn connect failure cache design to the spec
Tor will remember TCP connection failures to relays, and if it gets a request to make another connection to that relay within that minute, it will abort the attempt. See tor#24767 (closed) (commit f29d1583, went into Tor 0.3.3.4-alpha) for details.
Anybody building a Tor relay will want to add this feature too, so we should specify it somewhere. I'm not sure what is the best spec file though. Maybe dos-spec.txt? Or maybe tor-spec.txt because it is actual fundamental behavior that SHOULD be done to have implemented Tor correctly?