fix: LaTeX rendering, Exercise C worked example, Python 3.11+ requirements, Sphinx docs

[Copilot]

Summary

Addresses all four items from the problem statement.

Changes

1 — Fix LaTeX rendering in notebook section "1) Mathematical intuition"

• All markdown cells in the demo notebook that used \(...\) / \[...\] LaTeX delimiters (cells 1 and 9) have been converted to the universally-supported $...$ / $$...$$ notation. Both classic Jupyter Notebook and JupyterLab render these via MathJax without additional configuration.

2 — Worked example for Exercise C (correlated predictors)

• Added two new cells (7c) immediately after the Exercise A sweep:
  • Markdown cell: Explains the identifiability challenge posed by correlated features and what to expect from each selection method.
  • Code cell: Builds 4 blocks of 5 features with within-block correlation ρ = 0.9, places a true signal in the first feature of each block, fits HorseshoeGLM with MCMC, then compares method='beta', 'lambda', and 'both' side-by-side (precision / recall / F1 table + bar-chart of inclusion probabilities). Ends with a plain-language discussion of why posterior mass spreads and when alternative priors may be preferable.

3 — Remove Python < 3.11 support

• pyproject.toml: Added requires-python = ">=3.11" and removed 3.9/3.10 classifiers.
• tox.ini: Updated envlist to py311, py312 and updated the [travis] mapping.
• .travis.yml: Updated test matrix to 3.11 and 3.12; updated the deployment python key to 3.11.

4 — Fix Sphinx doc build errors

• docs/conf.py: Added sys.modules pre-mocking of heavy C-extension dependencies (JAX, NumPyro, ArviZ, scikit-learn, etc.) using unittest.mock.MagicMock so that bayesian_feature_selection can be imported without those packages being installed. Also added autodoc_mock_imports for the same list so that sphinx.ext.autodoc resolves type annotations correctly. The Sphinx build now completes successfully (build succeeded, 0 warnings, 0 errors).