handle the case of duplicate slaves better

[dustin]

For various reasons, we at Zmanda often see two slaves running on the same machine. Needless to say, these slaves continually boot one another off the master, wasting lots of resources.

The current buildmaster code (master.buidlbot.buildslave.AbstractBuildSlave.attached) doesn't handle this very well.

I think that the best approach is to ping the old slave and reject the new slave. If the old slave really is gone, then the ping should time out, and the new slave will eventually reconnect without any conflict. If the old slave isn't gone, then the new slave will just bang away endlessly at the master until someone reads the logs and notices -- which is better than interrupting builds!

[Dustin J. Mitchell]

reject the *new* slave when a dupe slave connects

This, coupled with a ping of the old slave, will keep double slaves from killing one another

Fixes #962

Changeset: 9b8170191deb587df4395fb44931071b1df32b25