Containers configured with runblocking should exit or give up blocking eventually

[hickeng]

Story
As a user I expect my containers to start successfully or fail to start with a helpful error message

Detail
We have runblocking to address the case where it's absolutely necessary to get all container output - however a container configured in this manner will block indefinitely until an attach attempt makes it to the container and unblocks the process launch.

Two items to address:

• eventually we have to conclude that we should launch anyway, or shutdown because the attach isn't coming Request canceled when waiting on container VM process to launch #4289
• irrespective of current state it should be possible to stop and then rm containers Containers can get into a state where the container cannot be deleted #4269

Related: #4296 #4269

Notes
We have to decide what the actual behaviour should be in the case that attach fails to start - which is more important, maintaining the succeed totally or fail, or having the process run and being able to grab the logs later.

The latter is a more tolerant approach and might be more suited to a production environment, but is a harder behaviour to document and maybe not intuitive.

[hickeng]

Implementing this would work around #5479 to some extent, although it should not be considered a fix.

[mdubya66]

Power cycle via the UI might be a workaround

[renmaosheng]

the items mentioned in the first description are all fixed. closing this.