Eats CPU and memory due to slighly misconfigured OOPSObserver

In txlongpoll, an observer is created by getRotatableLogFileObserver()
and passed as the fallback parameter to OOPSObserver. The docstring
for OOPSObserver.__init__ says this is the right thing to do:

        :param fallback: If supplied, a LogObserver to pass
            non-failure log events to, and to inform (as non-failures)
            when an OOPS has occurred.

However, it actually expects a callable, but this does not cause an
error until later.

I'm not quite sure why it doesn't happen every time, but what can
happen is that the txlongpoll process spins tightly and comsumes more
and more memory. I've attached an example log that is recorded.

The following diff, I think, will resolve this issue in txlongpoll:

{{{
=== modified file 'twisted/plugins/txlongpoll.py'
--- twisted/plugins/txlongpoll.py 2011-09-30 10:46:17 +0000
+++ twisted/plugins/txlongpoll.py 2011-10-06 19:48:36 +0000
@@ -67,7 +67,7 @@
         repo = DateDirRepo(options["oops-dir"], options["oops-prefix"])
         config.publishers.append(defer_publisher(repo.publish))

- observer = OOPSObserver(config, logfile)
+ observer = OOPSObserver(config, logfile.emit)
     addObserver(observer.emit)
}}}

To avoid this kind of issue in the future, I propose doing something
like the following in python-oops-twisted:

{{{
=== modified file 'oops_twisted/log.py'
--- oops_twisted/log.py 2011-09-28 09:00:46 +0000
+++ oops_twisted/log.py 2011-10-06 19:57:22 +0000
@@ -41,6 +41,7 @@
             occurred.
         """
         self.config = config
+ assert fallback is None or callable(fallback)
         self.fallback = fallback

     def emit(self, eventDict):
}}}