Fix QUIC crash on ConnectionClose, protected behind renamed flag

Fix QUIC crash on ConnectionClose, protected behind renamed flag
FLAGS_close_quic_connection_unfinished_streams_2 (defaulted to false).

With this change we avoid trying to close the connection while driving
the QUIC internal server response pipeline, as the call to
SendConnectionClose now happens during PostProcessAfterData, which is
only ever called after receiving new data. With a hacky
client/QUIC internal server combo (sending > 100 requests, but never
sending FIN/RST) I've verified that with this change we don't hit the
DFATAL from the attached bug.

Still not entirely sure why Chrome would trigger this, b/17402744 is
open to investigate why it ever opens more than kMaxStreams.

Merge internal change: 75541290

R=rch@chromium.org, rjshade@chromium.org

[ramant (doing other things), 2014-09-15 20:09:26]

https://codereview.chromium.org/575593002/diff/1/net/quic/quic_flags.cc
File net/quic/quic_flags.cc (right):

https://codereview.chromium.org/575593002/diff/1/net/quic/quic_flags.cc#newco...
net/quic/quic_flags.cc:33: bool FLAGS_close_quic_connection_unfinished_streams_2
= false;
rch/rjshade: should we enable this flag in chrome? thanks.

[Ryan Hamilton, 2014-09-15 21:45:55]

lgtm

https://codereview.chromium.org/575593002/diff/1/net/quic/quic_flags.cc
File net/quic/quic_flags.cc (right):

https://codereview.chromium.org/575593002/diff/1/net/quic/quic_flags.cc#newco...
net/quic/quic_flags.cc:33: bool FLAGS_close_quic_connection_unfinished_streams_2
= false;
On 2014/09/15 20:09:26, ramant wrote:
> rch/rjshade: should we enable this flag in chrome? thanks.

My recommendation would be to let this run server-side for a while to confirm
that it's safe, since we're not in any hurry to see this behavior in Chrome.
rjshade: cool with you?

[Robbie Shade, 2014-09-15 21:48:09]

On 2014/09/15 21:45:55, Ryan Hamilton wrote:
> rjshade: cool with you?

Undoubtedly.

lgtm.

[ramant (doing other things), 2014-09-15 21:49:15]

https://codereview.chromium.org/575593002/diff/1/net/quic/quic_flags.cc
File net/quic/quic_flags.cc (right):

https://codereview.chromium.org/575593002/diff/1/net/quic/quic_flags.cc#newco...
net/quic/quic_flags.cc:33: bool FLAGS_close_quic_connection_unfinished_streams_2
= false;
On 2014/09/15 21:45:55, Ryan Hamilton wrote:
> On 2014/09/15 20:09:26, ramant wrote:
> > rch/rjshade: should we enable this flag in chrome? thanks.
> 
> My recommendation would be to let this run server-side for a while to confirm
> that it's safe, since we're not in any hurry to see this behavior in Chrome.
> rjshade: cool with you?

Cool. thanks