Think about how subgraphs should be implemented

[mlaveaux]

Currently, in many places where a graph is represented: LTS, PG and Graph traits it's convenient that the vertices are numbered 0 to num_of_vertices(). For example the indexed partition works this way, since indexing into a table is very cheap.

However, for a subgraph, which is simply another struct that implements the same trait, we can either decide to ensure that all vertices are remapped to dense vertex indices. However, this requires keeping track of two mappings from and to the old and new indices. Alternatively, we keep num_of_vertices() the same for the subgraph, but ensure that only indices of vertices part of the subgraph are returned with their original vertex.

[mlaveaux]

In Zielonka's recursive implementation it could be interesting to consider three different types classes of subgames in the recursive calls. Initially, we pass the full game (instead of making the full vertex set), then we have the bit vectors for subgames as we have now. However, if the games are small enough it would be interesting to use a HashSet to represents the subgames.