Add HSCIC-CPT conditional independence test

[felix-laumann]

Summary

This pull request adds HSCIC-CPT as a conditional independence test and exposes it through causal-learn's existing CIT and PC interfaces.

• implements the Hilbert-Schmidt conditional independence criterion (HSCIC)
• calibrates HSCIC with the exchangeable conditional permutation test (CPT) sampler
• uses a held-out training/evaluation split and selects the HSCIC ridge on training data only
• supports the built-in linear-Gaussian conditional-density model or a user-supplied conditional-density callback
• evaluates conditional queries in both directions and combines the directional p-values with a Bonferroni-adjusted minimum, avoiding dependence on arbitrary column ordering
• adds API documentation, references, and dedicated tests

Implementation fidelity

The CPT transition uses likelihood-matrix orientation, identity start, shared Markov-chain centre, disjoint-pair proposals, swap log-odds, and numerical truncation. Kernel bandwidth selection follows lower-median convention, including duplicated observations.

The lower-level HSCICCPT class remains directional because CPT resamples one variable conditionally. The public causal-discovery wrapper makes the test symmetric by testing both directions.

Validation

• python -m unittest tests.TestCIT_HSCICCPT — 13 tests passed
• python -m pytest tests/TestCIT_HSCICCPT.py -q — 13 tests passed
• PC integration smoke test passed
• Sphinx HTML documentation build completed successfully
• compileall, codespell, and git diff --check passed

References

• HSCIC: Park and Muandet (2020), https://proceedings.neurips.cc/paper/2020/hash/f340f1b1f65b6df5b5e3f94d95b11daf-Abstract.html
• CPT: Berrett et al. (2020), https://academic.oup.com/jrsssb/article/82/1/175/7056014
• HSCIC-CPT: Laumann et al. (2023), https://www.mdpi.com/1099-4300/25/12/1597