Better handling of --stack-name filtering

Using the create or upsert commands combined with the --stack-name filter on a stack that has a stack dependency will not work. A possible solution could be to dynamically load stack outputs from all relevant stacks.

Since we will have already created a dependency graph beforehand, determining which stacks' outputs are needed would simply be a matter of reversing the direction of all edges and determining the stack's neighbours (alternatively, look at the source node of all incoming edges).

The inn function from fgl could be useful.