docs: correct docstrings across batch, multivariate, regression, univariate

[kgdunn]

Summary

Follow-up docstring audit after #495. Six small drifts between what the docstring says and what the code does, plus one tiny code fix to make two stubs behave the way their docstrings claim.

• src/process_improve/batch/data_input.py (line 174): melted_to_wide unconditionally returned {} while its docstring simply said "Convert aligned melted data to wide format." Docstring now says "Not yet implemented (returns an empty dict); see also dict_to_wide."
• src/process_improve/batch/data_input.py (lines 194, 200): wide_to_melted and wide_to_dict said "Not yet implemented" while silently returning empty structures. wide_to_dict also took no arguments, so any caller passing a DataFrame got a TypeError. Both now accept the expected input and raise NotImplementedError so the behaviour matches the docs. The two stub tests were updated to pytest.raises(NotImplementedError).
• src/process_improve/multivariate/_pca.py (lines 220-233, 271-280): PCA docstring now documents n_components : int or None, and __init__ defaults n_components to None; this matches _parameter_constraints ([int, None]) and fit(), which already clamped None to min(n_samples, n_features).
• src/process_improve/multivariate/_pca.py (line 302): the fitting_info_ inline comment had algorithm-to-shape backwards. Corrected to "per-component arrays (SVD/NIPALS) or scalar totals (TSR)."
• src/process_improve/regression/_robust_regression.py (line 135): the conf_intervals entry in robust_regression's Returns block described it as K rows x 2 columns, but the function only supports single-predictor regression. Rewritten as 1 x 2 (lower, upper) confidence interval for the single slope coefficient. The K-row wording on the sibling multiple_linear_regression (line 325) was left untouched.
• src/process_improve/univariate/metrics.py (lines 63-70): t_value_cdf doctests used np.inf without importing numpy into the doctest namespace. Replaced with float("inf") / float("-inf"), which needs no import and matches the surrounding style. t_value didn't actually use np.inf in its doctest code (only in the :math: prose), so it was left alone.

Version bumped 1.66.2 -> 1.66.3 (PATCH, docs). CITATION.cff synced to the new version and today's date (2026-08-21). New Documentation bullet added under ## [Unreleased] in CHANGELOG.md.

Test plan

• uv run ruff check src/process_improve/... tests/batch/test_data_input.py - passes.
• uv run ruff format --check src/process_improve/... tests/batch/test_data_input.py - 5 files already formatted.
• uv run pytest tests/batch/test_data_input.py --no-cov - 10 passed.
• uv run pytest tests/test_multivariate.py tests/test_regression.py tests/test_univariate.py tests/batch/ --no-cov - 302 passed, 2 skipped. The 4 errors seen are pre-existing ImportError for the optional scikit-fda/pyDOE3 extras and are unrelated to this change.

Checklist

• Version bumped in pyproject.toml (PATCH, docs-only)
• Tests added or updated where relevant (two stub tests re-pointed at pytest.raises(NotImplementedError))
• ruff check . passes
• CHANGELOG.md updated
• CITATION.cff version and date synced

---

Generated by Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!