Remember which directory caches give us unusable consensuses.

If we are looking for a usable consensus and our directory has given us one that's far in the past, we should not try that same directory again right away!

Unfortunately, that seems to be what our current code will do. One option here is to kill the circuit to that directory and hope that we pick a better one at random. Another one is to be a bit more thorough about somehow marking that directory as misbehaving, perhaps in the tor-netdir code.

Found while investigating #329 (closed).