Nested stack updates are still serialised in convergence

[Zane Bitter]

One of the goals for convergence is that we deserialise stack updates by allowing simultaneous traversals of the resource graph. This is accomplished by locking at the Resource level instead of at the Stack level.

A consequence of this is that when updating a nested stack, we hold a lock on the parent resource - which effectively functions as a lock at the stack level.

While we still get the benefit of always accepting the user's update request (i.e. never responding with an operation in progress error), we don't get the benefit of being able to act immediately on new information.

Since a stack can be updated at any time, we should find a way to force an old traversal to release the lock on an in-progress nested stack resource so that we can immediately begin the new update. This would probably take the form of cancelling the update of that particular resource (*without* rollback).