Add support for simulating failures in the cluster.

For example, during a test run, at a given iteration simulate that a component (e.g., ovsdb-server) crashes.

One way of currently doing this is to use the `ext_cmd` support in ovn-heater.  It's generic enough so it might turn out that writing some samples about how to simulate component failures at different times is enough for closing this issue.