Add option to limit the number of targets in a PodDisruptor

[pablochacin]

Presently, when creating a PodDisruptor, all pods that match the pod selector become targets of the subsequent fault injection(s). However, there are cases in which it would be convenient to limit the fault injection to a subset of pods. For example

• To simulate faulty instances in a deployment (typical of a canary release or rolling deployment)
• To simulate an under-performing instance and validate if requests are correctly load-balanced

The initial idea, aligned with the design of the selector API could be to add a limit attribute to the pod selector, that allows defining a maximum number of pods to be selected. For example:

selector = {
    select: { labels: {...} }
    limit: 1
}

However, in some cases, a percentage of the pods could be more convenient (for example, when the number of replicas varies):

selector = {
    select: { labels: {...} }
    limit: '20%'
}

One open question in this proposal is how to limit the number of targets for a ServiceDisruptor, as it does not consider a selector.

Note: Having arguments that allow multiple types of values is not uncommon in Kubernetes (for example, disruptor budget) but this is not necessarily a requirement for the API, just an example of the concept.