On Error Target for a stack

I think we should add a 'deploy_error' to the targets of a stack.

This would be treated like a normal stack, but only ran if a deployment failed.  it would be handed information about the failure (as much as possible).

This would let people rollback or do other processes automatically.
