Fix all_node_cuts output for complete graphs

[navyagarwal]

Closes #6533

I modified the special condition for complete graphs in all_node_cuts to return an empty generator and also changed the test cases accordingly.

[dschult]

Great! This should be merged after v3.2 is released and included in the release (along with removal of the deprecation warnings and conftest.py entires) of v3.3.

[MridulS]

@navyagarwal if you still have the bandwidth could you add the deprecation warnings so we can push this as part of v3.3. Thanks!

[dschult]

Based on my comments in #6533 I think there is more needed here.
And I'm not seeing now what would be deprecated.
The main problem is:

G=nx.complete_graph(5)
G.remove_edges_from([(0, 1), (3, 4)])
print(list(nx.all_node_cuts(G)))  # returns [{0, 1, 2}]

returns [{0, 1, 2}] when it should also include {2, 3, 4}.
Similar looking case works correctly:

G = nx.complete_graph(5)
G.remove_edges_from([(1, 2), (0, 4)])
print(list(nx.all_node_cuts(G))) # returns [{1, 2, 3}, {0, 3, 4}]

[dschult]

Yikes -- that took a bit. But I think I found an error in nx.all_node_cuts. Luckily it wasn't in the heart of the algorithm, but in the setup for which nodes get considered. A change from removing the potential node_cut set X to removing the currently considered node in that set {x} did the trick. I was following along with the article and it says "if x_i != v_j and v_j is not adjacent to x_i". It does not say that v_j should not be in the set X.

Anyway, it now works! With the fix already here for complete graphs, we have solved all the issues raised in #6533.

Ready for Review

[MridulS]

thanks @dschult @navyagarwal !