support scaling of the cluster inside the runner

I am working on a cluster now where I can dynamically scale the cluster using SLURM + jobqueue.

There the Client immediately returns and can have only a fraction of the number of cores that you reserved. This means that if you start a runner at that moment, it will not use all the resources later on.

With this simple change that is fixed.

@jbweston and @anton-akhmerov what do you think?