Implement a directed graph and its iterator

This implementation comes initially from the one in gpr2 (GPR2.View_Ids.DAGs) and
from e3-core (src/e3/collection/dag.py).

Each node in the graph is identified by a node ID and can have other nodes
as predecessors. Acyclic graphs can be iterated with an iterator that can
be external or internal to the graph.

Each graph node supports three different states during an iteration:
non-visited, visiting and visited:

* Non_Visited : The node has not been iterated.
* Visiting    : The node has been iterated but it has not been processed by
                the graph iterator caller. Use the "Complete" procedure to go
                from the "Visiting" to the "Visited" state.
* Visited     : The node has been iterated and processed.

However, it is also possible to only keep the non-visited and visited states
to simplify the process.

Nodes can be added or their predecessors updated during an iteration.
Notice that a node can not be removed, for simplicity and efficiency.

Co-authored-by: Nicolas Roche <roche@adacore.com>

Author: Jicquel