feat: Add topological Betti-1 anomaly detection and fix connected components

[teerthsharma]

Summary

This PR introduces a lightweight topological defense layer for detecting structurally anomalous graph inputs by analysing their first Betti number (beta_1), which counts independent cycles. It also fixes the broken connected-components placeholder in global_graph_topology.py.

Changes

• dgf/src/analyse/topology/betti_defense.py (new):
  • calculate_betti_1(graph, schema) -- returns beta_1 for homogeneous graphs.
  • is_anomalous(graph, schema, expected_max_betti) -- boolean flag for anomalous graphs.
  • _connected_components(adjacency, num_nodes) -- union-find implementation for CC labels.
• dgf/src/analyse/topology/betti_defense_test.py (new): 10 tests covering trees, cycles, cliques, self-loops, empty graphs, and heterogeneous-graph rejection.
• dgf/src/analyse/topology/global_graph_topology.py:
  • Replaced the empty cc = np.array([]) placeholder with a real union-find call.
  • Added betti_1 field to GlobalGraphTopology.
• dgf/src/analyse/topology/BUILD:
  • Added betti_defense py_library target.
  • Updated global_graph_topology deps.

Background

I have also applied topology in machine learning, and I currently have an open (not yet closed or merged) contribution in deepmind/penzai. I hope this module proves useful for the DGF community.

Thank you for your time.