Fix QUIC alternate protocol behavior so that that main job does not wait

net/quic/quic_client_session.cc

Index: net/quic/quic_client_session.cc
diff --git a/net/quic/quic_client_session.cc b/net/quic/quic_client_session.cc
index d5bb42ef8f62d8ff4303d87e900d861dec6813a5..8053de5565c95f3227b3f5621294652ca5bbbdc8 100644
--- a/net/quic/quic_client_session.cc
+++ b/net/quic/quic_client_session.cc
@@ -70,7 +70,6 @@ QuicClientSession::QuicClientSession(
 }
 
 QuicClientSession::~QuicClientSession() {
-  DCHECK(callback_.is_null());

[willchan no longer on Chromium, 2013/06/21 23:55:45]

Why did this change?

[Ryan Hamilton, 2013/06/22 00:00:36]

Ah, right.  This was a check that raman added when we ported the code to
Chromium.  At the time, we though that this check was a good sanity check
because it should never happen that we are destroyed while we are connecting. 
However, this turns out not to be correct.  If we shutdown during a connection,
we hit this.  In the new test I added, that's exactly what happens.  The
QuicClientSession is still connecting (and hence the callback is set) when the
stack shuts down.

[willchan no longer on Chromium, 2013/06/22 00:14:33]

Great, sounds good.

   connection()->set_debug_visitor(NULL);
   net_log_.EndEvent(NetLog::TYPE_QUIC_SESSION);