Add ER-based z-score normalization to metrics

Implement a decorator to return z-scores vs directed G(n,m) ER graphs (igraph.Graph.Erdos_Renyi). Normalization is binary-only; weighted normalization is not supported. API: normalize: bool | None (default None), n_random: int = 20, random_seed: int | None. If normalize is False/None, metric behaviour is unchanged and passing normalization parameters raises an error. Update Sphinx docstrings, guide with rationale and examples (citing Zanin et al., 2018), and add comprehensive tests.