Adapt pod killing and cleanup for generic container runtime

[yujuhong]

This change removes docker-specifc code in killUnwantedPods. It
also instructs the cleanup code to move away from interacting with
containers directly. They should always deal with the pod-level
abstraction if at all possible.

[yujuhong]

/cc @vmarmol, @yifan-gu

To avoid inspecting individual containers, I had to use containerManager.GetPods. This bypasses the runtime cache and may affect performance (?). I am doubtful that this operation is so expensive that we need to use the runtime cache at all times (since this is called only once per sync). But I suppose the cache was added for a reason.... If we absolutely want to avoid calling this function, I could modify the container runtime interface to support a function similar to GetRunningContainers to take container ids, inspect them, and return running pods. This doesn't seem to play well with rkt though.

[vmarmol]

LGTM. Nice! We take out a bunch of code and end up with something more generic :)

I agree that the performance impact is probably minimal since this is only a single extra call per sync loop. If anything, with time this operation will become async too.

[yifan-gu]

Why we pass true here? I remember true is for getting all pods including dead ones?

[yujuhong]

Good catch. I meant to say false...Thanks!

[yifan-gu]

Great! This is what I want to do! Thanks @yujuhong !

[yujuhong]

Rebased. Thanks!

[yifan-gu]

LGTM