We should support running spot instances on AWS.
Things to consider:
- should it be per-instance or global per-cluster?
- how do we get notified of an impending termination?
- how should we handle the termination?
- what does the compute node need to do?
- what does the management node need to do?