Verify path bias code plays well with optimistic data
In connection_ap_handshake_socks_reply() and some other places, we mark circuits as having successfully carried streams if we get certain end reason codes back. With optimistic data enabled, we may actually get these same end reason codes back even if the circuit/stream was damaged due to tagging after accepting an optimistic data stream attempt.
I probably need to instrument a tor client running with optimistic data to verify these codepaths can't happen.