Update ANM to be compatible with the latest version of sklearn

[ErdunGAO]

Updates

• Clean some useless codes (mainly about including paths).
• Update ANM to be compatible with the latest version of scikit-learn. (PR 22199 into scikit-learn ([BUG Fix covariance and stdev shape in GPR with normalize_y scikit-learn/scikit-learn#22199]) changed the shape of the array returned by GaussianProcessRegressor.) Thanks to @Jonathan-Salisbury for finding this issue (PR 112 to causal-learn [ANM incompatible with scikit-learn >= 1.1.3 #112]).
  1. Change the shape of the output of GaussianProcessRegressor.
  2. Change the value dataset_p_value_forward from 0.14736 to 0.14773.

Test check

python -m unittest tests.TestANM # should pass