update session state transitions from distributed servers

```
The RUNTIME and SHUTDOWN states should not be entered by the master session
until the slave sessions on distributed emulation servers have entered those
states. Instead of keeping node counts, we can just wait for the slave server
to keep track of its nodes and enter the state.

Currently, RUNTIME is reached by snooping node STR messages, the ones that
cause the red square node highlight to turn green. SHUTDOWN is currently
declared possibly before all of the slaves have been shutdown.

Once SHUTDOWN is reached, the Broker.recvloop() should be terminated.
```

Original issue reported on code.google.com by `ahrenh...@gmail.com` on 23 Jul 2012 at 10:12

Start by tracing the master session's RUNTIME and SHUTDOWN transitions, including the node STR handling described in the issue, and inspect Broker.recvloop(). Follow how slave servers track their nodes and report state. Done means the master enters each state only after all distributed slave sessions do, and Broker.recvloop() terminates once SHUTDOWN is reached.

Written by the indexing model from the issue text.