Seed the delta SVM regression test - #688

Closes #687.

Request the existing set_seed fixture in test_regression_delta_svm. The raw bootstrap in delta.bias_reduced_delta uses NumPy's legacy global RNG, so the unseeded test depended on random draws consumed by preceding tests.

This changes only the test's fixture dependency and adds an explanatory comment. The fixture's existing seed (123456), estimator, bootstrap count, expected values, and tolerances are unchanged. It is separate from the discrepancy-sampler fix in #685.

Validation on the current main base (aa2c554):

• Python 3.10.20 / NumPy 2.2.6 / SciPy 1.15.3: 17 targeted tests passed.
• Python 3.14.6 / NumPy 2.4.3 / SciPy 1.17.1: the same 17 tests passed.
• Applying the fixture after two different ambient seeds produced identical estimates in the Python 3.14 environment.
• Black 24.2.0, Ruff 0.3.0, and git diff --check passed.

python -m pytest -q tests/test_analyze_delta.py tests/test_regression.py::test_regression_delta_svm

These are targeted local checks, not a full-suite or cross-platform CI pass.