Add n_components option to BayesianPCA

[joon0390]

Summary

This PR adds an n_components option to the BayesianPCA class, allowing users to directly specify the latent dimensionality instead of being forced to use d-1.

Changes

• BayesianPCA.__init__ now accepts n_components (default: None).
• If n_components is not provided, it falls back to the original behavior (q = d - 1).
• Added fallback handling when eff_dim is empty (use all components).
• Updated docstring for clarity.

Motivation

Currently, BayesianPCA always uses d-1 latent dimensions, which is impractical for high-dimensional datasets.
This update improves usability by letting researchers control the number of latent dimensions (e.g., 10, 20, …), similar to scikit-learn's PCA API.

Validation

• Verified on real fMRI FC dataset (ROI=300): stable latent representations produced.

Checklist

• Code runs without errors
• Works with both default (d-1) and user-specified latent dimensions
• Eff-dim fallback tested