Add fast pytest for skpro FunctionalCPD in FunctionalBN refactor

[daehyun99]

Motivation

• Provide CI coverage for the skpro integration path in FunctionalCPD without running expensive Bayesian training.
• Ensure FunctionalBayesianNetwork correctly invokes the external-model adapter for tag='skpro' and passes the expected input shapes.

Description

• Added pytest import and a new test test_fit_with_bayesian_linear_regressor_uses_fast_mocked_fit to pgmpy/tests/test_models/test_FunctionalBayesianNetwork_Refactor.py that exercises the skpro CPD path with a small synthetic dataset.
• The test uses pytest.importorskip for skpro and pymc_marketing and monkeypatch to replace BayesianLinearRegressor.fit with a fast mock that records X/y shapes and column names.
• The test constructs a FunctionalBN with mixed CPD tags (tabular, skpro, linear), calls model.fit(data), and asserts the adapter supplied a constant-design matrix (['_const_']) of shape (n_samples, 1) and the expected y shape.

Testing

• Ran pytest -q pgmpy/tests/test_models/test_FunctionalBayesianNetwork_Refactor.py which completed with 2 passed, 1 skipped.
• Attempted pre-commit run --all-files but it failed in the environment with pre-commit: command not found.

---

Codex Task

[gemini-code-assist]

Code Review

This pull request introduces a new test case, test_fit_with_bayesian_linear_regressor_uses_fast_mocked_fit, to verify the integration of FunctionalBayesianNetwork with skpro's BayesianLinearRegressor using a mocked fit method. The review feedback recommends using monkeypatch.setitem when modifying the _tags attribute of BayesianLinearRegressor to prevent side effects from persisting across other tests.