Fix ChromeFrame referrer test which was flaky due to incorrect http header pa...

Fix ChromeFrame referrer test which was flaky due to incorrect http header parsing. To pass the referrer
to Chrome during navigation we were converting normalized HTTP headers which are terminated with \r\n to
raw headers and attempting to parse them. Converting to raw headers basically replaces \r\n with the NULL
terminator which causes the parsing to fail as a result of which we fail to retrieve the referrer.

Fix is to not convert the headers to raw and just parse the orignal headers.

The other change is to only flag success or failure in the referrer frame test if we are running in Chrome.
This page loads initially in IE as well due to the vagaries of the http equiv patch. Flagging an error
if we don't find a referrer basically causes the test to fail.

This should fix http://code.google.com/p/chromium/issues/detail?id=34812

Will remove the flaky attribute on the ChromeFrame Referrer test in a subsequent CL if all goes well.

Bug=34812
Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=42372

[tommi (sloooow) - chröme, 2010-03-23 16:11:59]

lgtm.

We need a test that verifies that scripts in CF pages never execute in mshtml :(

Also, if you could check if UrlmonUrlRequest might have problems too that'd be
great (see below)

http://codereview.chromium.org/1128013/diff/1/3
File chrome_frame/urlmon_moniker.cc (right):

http://codereview.chromium.org/1128013/diff/1/3#newcode36
chrome_frame/urlmon_moniker.cc:36: net::HttpUtil::HeadersIterator
it(raw_headers_utf8.begin(),
Doh! We're using AssembleRawHeaders in UrlmonUrlRequest::OnResponse as well.  Do
you think we might have this issue there too?

[tommi (sloooow) - chröme, 2010-03-23 19:01:56]

lgtm.