BackoffEntry: Add the option to always use a delay.

remoting/host/chromoting_host.cc

Index: remoting/host/chromoting_host.cc
===================================================================
--- remoting/host/chromoting_host.cc	(revision 131823)
+++ remoting/host/chromoting_host.cc	(working copy)
@@ -54,6 +54,9 @@
   // Time to keep an entry from being discarded even when it
   // has no significant state, -1 to never discard.
   -1,
+
+  // Don't use initial delay unless the last request was an error.
+  false,

[Jói, 2012/04/24 20:51:20]

Did you mean to set this to true and update the comment?  If not, what was the
motivation for this change?  (I like the change anyway)

[mmenke, 2012/04/24 21:05:25]

The motivation is my unlanded captive portal detection CL.  Nothing to do with
this file.  We want to repeatedly request the same file for checking for captive
portals, but since a lot of things that trigger the captive portal check may
happen in a row (A bunch of network errors), we want a pretty significant
timeout between checks, normally (Currently 10 minutes, when not behind a
captive portal, though we may play with that significantly, and maybe reset the
failure count on network changes).

If we get the same result too often, we then want to start exponential backoff,
so we don't hammer servers when the check is consistently getting triggered, but
our captive portal state isn't changing.

 };
 
 }  // namespace