Removed the call to purge() on the timer that kills queries that over…

…run their timeout.

We were calling it for every task, and it's linear in the number of tasks, and synchronized
on the timer's monitor, so it was effectively quadratic on the number of pending tasks.

It turns out that we don't need to call purge() at all anyway...

src/main/scala/com/twitter/querulous/Timeout.scala

@@ -23,7 +23,9 @@ object Timeout {
     } finally {
       task map { t =>
         t.cancel()
-        timer.purge()
+        // TODO(benjy): Timer is not optimized to deal with large numbers of cancellations: it releases and reacquires its monitor
+        // on every task, cancelled or not, when it could quickly skip over all cancelled tasks in a single monitor region.
+        // This may not be a problem, but it's something to be aware of.
       }
       if (cancelled) throw new TimeoutException
     }