Implement the WindowsLogoffRace finch experiment.

chrome/browser/chrome_browser_field_trials.cc

Index: chrome/browser/chrome_browser_field_trials.cc
diff --git a/chrome/browser/chrome_browser_field_trials.cc b/chrome/browser/chrome_browser_field_trials.cc
index 655cf4564411ecff8fc0dc58aa88ac0a96d41633..b325e87ffb558a56882c0b244e754ad1049491ac 100644
--- a/chrome/browser/chrome_browser_field_trials.cc
+++ b/chrome/browser/chrome_browser_field_trials.cc
@@ -57,6 +57,7 @@ void ChromeBrowserFieldTrials::InstantiateDynamicTrials() {
   base::FieldTrialList::FindValue("InstantDummy");
   base::FieldTrialList::FindValue("InstantChannel");
   base::FieldTrialList::FindValue("Test0PercentDefault");
+  base::FieldTrialList::FindValue("WindowsLogoffRace");

[Alexei Svitkine (slow), 2014/07/14 18:55:02]

I don't think this is necessary since your experiment runs entirely in the
browser process.

[Sigurður Ásgeirsson, 2014/07/14 20:31:38]

I don't know how this machinery works, but the comment above leads me to believe
that my experiment will never be "used" until logoff. The metric I'm after will
however be reported right after launch, e.g. the unclean shutdown count.

Are you sure I'm good without this?
I guess I'm asking whether all UMA reports contain the experiment groups that
client is enlisted in?

[Alexei Svitkine (slow), 2014/07/14 21:40:32]

Having the trial start in "active" state makes sense. Still, you don't need this
line as that can be done from the server-side with a starts_active:true flag.

   // The following trials are used from renderer process.
   // Mark here so they will be sync-ed.
   base::FieldTrialList::FindValue("CLD1VsCLD2");