Implement quadratic regularization for TRF (Issue #16)

[Hugo-W]

Summary

Implements quadratic regularization for TRF estimation as described in Issue #16.

Changes Made

solvers.py

• Added svd_solver() support for quadratic regularization matrix M (solves (A+M)x = b using eigendecomposition)
• Added create_laplacian_matrix() helper function for smoothness constraints
• Added create_quadratic_regularizer() factory function
• Updated _svd_regress() to accept and pass M parameter

models.py

• Added quadratic_reg parameter to TRFEstimator.__init__() (accepts None, numpy array, or string 'smoothness'/'laplacian')
• Updated fit() and _fitlists() to build and pass M matrix to _svd_regress()
• Added standardized_coef_ attribute with standardized beta coefficients
• Block-diagonal Laplacian matrix creation for per-feature smoothness constraints

Usage Example

# Using smoothness constraint (Laplacian)
trf = TRFEstimator(tmin=-0.5, tmax=1.2, srate=125, 
                    quadratic_reg='smoothness')
trf.fit(X, y)

# Or with custom matrix
import numpy as np
M = np.eye(n_lags * n_feats) * 0.1  # Custom regularization
trf = TRFEstimator(tmin=-0.5, tmax=1.2, srate=125, 
                    quadratic_reg=M)
trf.fit(X, y)

# Access standardized coefficients
print(trf.standardized_coef_)

Testing

• Basic functionality with synthetic data
• Verify smoothness constraints work as expected
• Backward compatibility (quadratic_reg=None should behave as before)

Related Issues

Closes #16