Coalesce function and state for coo graphs

[dferre97]

Added a coalesce method and an is_coalesced flag to keep track of whether a COO graph is coalesced, which means:

• there are no duplicated coordinates in the indices (i.e, no multi-edges)
• the source and target indices are ordered lexicographically (by source, then by target)

Some operations (e.g., sparse matrix multiplication in forward pass) could be implemented more efficiently if a COO graph is coalesced.

[CarloLucibello]

can you test that operations like add_edges returning a new graph set coalesce to false?
Also we should add an expose an is_coalesced(g) method just returning g.is_coalesced.