Cluster run should return exit code

[jafreck]

Currently the cluster run return value is a mess, especially in the SDK:

The SDK returns a tuple of (node_id, output) where output can be a string representing the stdout of the command, or an exception in the event of a failure.

What should be returned is a dictionary:

{ nodes:
    node_id: {
        output: _
        exception: _
        exit_code: _
    }
}

Both the CLI and SDK should then return an exit code representing the success of the command where success means all nodes exited with a 0 exit code.