Run the post script _before_ killing the containers

[kegsay]

It's not very useful having the container ID otherwise..

[S7evinK]

Maybe should have added the reason why it is like this here as well when @kegsay asked for a review: The current approach makes sure that e.g. Dendrite is shutdown cleanly before attempting to pull out coverage data, which is only flushed out to disk on a clean shutdown (that's also why complementRuntime.ContainerKillFunc is a thing). Moving this up makes this now unusable for this use case.

I understand why it may be moved up, but this breaks coverage reporting for Dendrite.

From my perspective, we need both use cases.