1.7.0

Added

• Stimulus feature extraction (pyeeg.features package): LLM-based
  features (surprisal, entropy, KL divergence), syntactic features (tree
  depth, opening, closing), TextGrid alignment, feature pipeline, and
  dimensionality reduction (PCA, ICA).
• Robust Cauchy-loss TRF fitting (issue #17) via IRLS and a SciPy
  nonlinear least-squares reference path.
• Weighted samples (WLS) in TRFEstimator (issue #17).
• Feature-specific ridge regularization (banded ridge, feature_alphas).
• Quadratic regularization for TRFs (issue #16) with intercept and
  block-ordering support.
• pyeeg.models subpackage: split monolithic pyeeg/models.py into
  pyeeg.models (trf.py, var.py) while preserving public import paths
  (issue #11).
• Simulation-based regression tests for TRF and CCA (issue #10).
• Full test suite for pyeeg/utils.py (issue #9).

Changed

• Stabilized and documented public API entry points and pyeeg.__all__
  (issues #4, #5, #6, #24).
• Consolidated duplicate _svd_regress implementations (issue #7) and
  merged root solver.py into pyeeg/solvers.py (issue #8).
• Unified project documentation; removed outdated TODO/NEXT_STEPS files,
  replaced with ROADMAP.md.

Fixed

• P-value underflow and __repr__ crash in TRFEstimator statistics
  (issue #25).
• Gammatone variable mismatch with C extension (issue #26).
• Deprecated lag_matrix call arguments (issue #27).