Aggressively GC dead containers

[pwittrock]

We currently keep dead containers around for several reasons:

1. To keep container logs and other data around
2. To keep a record of the container state

• It is possible for the container to be GCed before we no longer need this state

We could address these in other ways by:

1. Storing records for dead containers elsewhere (perhaps the api server?)
2. Rotating log data to a permanent location (fluentd?)

This would allow us to immediately GC containers once their state and logs had been exported. It would also ensure that we did not lose state or log data due to premature garbage collection.

[dalanlan]

/cc @dalanlan

[dalanlan]

Just to mention, the logs stored elsewhere could be safely removed some time actually (those logs of a container corpse within a deleted pod, for example). The logs could take up space after all.

[yujuhong]

Closing this in favor of #26923