Inteference channel graph generation

[kevin-delmas]

There are two notable differences between the theoritical and implemented interference channel graph:

• all pairs of nodes used by transactions are not linked, only the linear paths are encoded (but I don't think this is a problem).
• unecessary nodes are considered, which is related to the FIXME on line 714 indicating that the implementation doesn't distinguish the origin of transactions when calculating channels (i.e., if two transactions from the same scenario share the same service, it's considered a potential interference channel that is not true. It's not severe, but it unnecessarily clutteres the graph.

[kevin-delmas]

@onera/phylog need to discuss if we add extra edges and remove spurious nodes or not

[Gaudeval]

An interesting case to consider is one where a single scenario has transactions which do not intersect. This implies distinct initiators, which is not quite desirable, but it may also cause the interference service graph to be disconnected.