Faster implementation for circuit_get_by_rend_token_and_purpose()

According to https://lists.torproject.org/pipermail/tor-relays/2013-September/002715.html , https://lists.torproject.org/pipermail/tor-relays/2013-September/002715.html can be 6% of a relay's runtime.

This is another function that does a linear search when a hashtable lookup would be more appropriate.

We'll need to be a little careful, since there's nothing preventing collisions here: An intro circuit and a rendezvous circuit can have the same "token" pretty easily.