Modification of the pod observation

[janLo]

This is mainly a RFC as I'm aware that it changes the current behavior quite a bit - and I'm fine to keep it as a fork. Still I'm curious what others think of these changes:

• I've removed the PodFailed branch from the pod observation as in my case, when I hit this branch, the reason and message was always empty. So I got a lot of events "Pod/xyz: " without any further helpful metadata. If there is a pod-change I always search for terminated containers.
• Every failed container causes a sentry event now and not only the first one. They will be grouped by a common fingerprint though.
• I've used the generateName field instead of uuid for the fingerprint (if the object has no owner) as this is the stable for a lot of solutions working with some kind of pod template (in my case Jenkins and Gitlab-Runner). If the generateName is not provided it uses the name in a mangled form like the sentry-kubernetes implementation did to allow better grouping.
• I took the message out of the fingerprints as its often containing very specific information like the container ids preventing any grouping.

[salzig]

@wichert what's your opionion on this?