Fix the replay of liveness and readiness probes

An assistant's Kubernetes manifest pointed both probes at a /healthz that checks a downstream API. When that API slowed down, the liveness probe failed and Kubernetes restarted the whole fleet in a loop. The two probes have opposite remedies: a failing liveness probe restarts the container, and a failing readiness probe only takes the pod out of the load balancer. The team now replays a probe's results to see which actions each one takes.

An assistant wrote simulate_probe_actions(kind, results, failure_threshold). It is in the editor. It has a bug. Find it and fix it.

Input and output:

• kind: "liveness" or "readiness".
• results: one probe result per tick, true for a success and false for a failure. Ticks are numbered from 0. The list can be empty.
• failure_threshold: how many failures in a row trigger an action, an integer. A value below 1 counts as 1.
• Return the actions in tick order, each as "<tick>:<action>", such as "3:restart".

Rules the tests check:

• Any other kind, such as "startup", returns an empty list. The match is case sensitive.
• Count failures in a row. Any success resets the count to 0, for both kinds.
• When the count reaches the threshold, reset it to 0, then act on that tick:
  • Liveness: add "<tick>:restart".
  • Readiness: if the pod is registered, mark it not registered and add "<tick>:deregister". If it is not registered, add nothing.
• Readiness only: the pod starts not registered. A success while it is not registered registers it and adds "<tick>:register". A success while it is registered adds nothing.
• Liveness never adds register or deregister.