Test validation of cyclic graph + local consts, doc failures

[acl-cqc]

• test_cyclic creates a cyclic DFG, and this validates :-(. We don't have anything in validation to look for this.
• test_local_const creates a graph with a LoadConstant node that is not in the causal cone of the Input. This passes validation, because the DfsPostOrder used in validate_children_dag currently traverses both ways along edges :-(, I think validation should fail here, and that the fix is in portgraph.
  • test_local_const then goes on to add the missing edge, and that still passes, but if we fixed the traversal to only go forwards, validate_children_dag would then fail as it wouldn't have visited the Const node (but would count it in the children); we'd then need the fix detailed inside validate_children_dag

I'd love to turn these into "expected-fail" tests but that looks quite hard, best plan appears to be to use cargo-nextest

[acl-cqc]

Closing in favour of #227 and #228