to fix bug where the burst stays high when limits
      are reduced.

      - Negative delay means run immediately and inline
      - Add methods to test class and run inline,
        to propagate errors to the console
      - Add javadoc for clients.config format
      - Use new methods for plugins

      - Big refactor of several classes for concurrent,
        elimination of several locks
      - Reduce max number of resent acks in a packet to
        lower overhead
      - Take incoming messages from the head of the queue,
        not sure why taking them from the tail "reduces latency"
      - Java 5 cleanup

      - Replace some locks with concurrent
      - Change job ID to a long so it won't wrap
      - Remove some unused stats
      - Java 5 and debug cleanup

      - Replace global counters with atomics
      - Use lockless shortcut methods to grant
        requests if we can satisfy immediately

      - Replace some locks with concurrent
      - Switch back to fast version of getPeersByCapability()
      - Java 5 cleanup

      - Remove some locks with concurrent

            to branch 'i2p.i2p.zzz.test' (head a4d67a357c36f4e94718bf237a7af96b8617a4a7)

      counter before the blocking write

     and '1d260b764d0fde4b8df67bdb23320b85e6944721'