Add Nadeau & Bengio corrected resampled t-test (closes #1075)

[azrabano23]

Closes #1075.

Within-session evaluation scores come from k-fold resamples whose training sets overlap, so the naive paired t-test underestimates the variance of the mean score difference and is overconfident. This PR adds the corrected resampled t-test of Nadeau & Bengio, which inflates the variance by the test/train ratio n2/n1 of each split:

t = ( (1/n) Σ xⱼ ) / sqrt( (1/n + n2/n1) σ̂² ), with t ~ Student(n−1) under H0.

What's included

• moabb.analysis.meta_analysis.compute_pvals_corrected_ttest(df, n_train, n_test, order=None): same interface and one-tailed k×k p-value convention as compute_pvals_wilcoxon / compute_pvals_perm, so it drops into the existing meta-analysis machinery (p-values kept strictly inside (0, 1) for Stouffer combination).
• Vectorized private core _corrected_resampled_ttest, following the _pairedttest_* style.
• 3 unit tests (synthetic data): exact agreement with the hand-computed formula and the algebraic shrinkage relation to scipy.stats.ttest_rel; conservativeness vs the uncorrected test with monotonicity in n2/n1; matrix properties, degenerate-variance edge cases, and input validation.
• API docs entry and changelog.

References: Nadeau & Bengio, Inference for the Generalization Error, NeurIPS 1999 / Machine Learning 52:239–281 (2003); Bouckaert & Frank, PAKDD 2004.

Scope note: per @bruAristimunha's comment on #1075, the natural follow-up is a splitter/evaluation hook that records n_train/n_test per fold and calls this test automatically from compute_dataset_statistics for within-session results — I kept this PR focused on the statistic itself and am happy to do the splitter integration as a follow-up if that split of work suits you.

[bruAristimunha]

many thanks for the contribution, can you update the tutorial too? 🙏🏽

[bruAristimunha]

many thanks @azrabano23 🙏🏽

[gcattan]

@bruAristimunha @azrabano23 I think we can make it a default in compute_dataset_statistic, and maybe deprecate the other methods.