Add random_state to CausalEstimator for reproducible bootstrap CIs and significance tests

[immu4989]

Closes #418
Closes #556

What

Bootstrap confidence intervals and significance tests in the shared CausalEstimator base class are non-deterministic, because they draw from the global numpy random state:

• _generate_bootstrap_estimates calls sklearn.utils.resample(...) with no random_state, so bootstrap CIs differ between runs.
• _test_significance_with_bootstrap calls np.random.permutation(...), so bootstrap p-values differ between runs.

This affects every estimator that uses bootstrap CIs or significance testing.

Changes

• Add a random_state parameter (int or np.random.RandomState, default None) to CausalEstimator.
• Add _get_random_state() and use it to seed both resample(...) and the permutation in the significance test.
• random_state=None preserves the existing (non-deterministic) behavior, so this is backward compatible.
• Only the resampling layers are seeded; a user-supplied propensity_score_model is left untouched.
• Add regression tests asserting that two runs with the same random_state produce identical confidence intervals and p-values, while different seeds differ.

Usage

estimate = model.estimate_effect(
    identified_estimand,
    method_name="backdoor.propensity_score_weighting",
    confidence_intervals=True,
    test_significance=True,
    method_params={"init_params": {"random_state": 42}},
)

Verification

• Same seed → identical CIs and p-values; different seeds → different CIs.
• Existing estimator tests pass; default behavior unchanged.

[amit-sharma]

thanks for this fix @immu4989 . Much appreciated

[amit-sharma]

@all-contributors please add @immu4989 for code

[allcontributors]

@amit-sharma

I've put up a pull request to add @immu4989! 🎉