v0.7.0

Added

• LDR (Lattice Dispersion Relation) polarizability (Draine & Goodman 1993) replacing Radiative Reaction model — reduces DDA-Mie discrepancy from 10-30% to ~1% at medium precision
• _compute_polarizability() function with per-solve LDR correction, supporting separate S parameters for x/y polarizations in depolarized mode
• Fourth precision level "ultra" (|m|kd ≤ 0.15) for very high accuracy DDA
• test_extract_dipoles.py rewritten with LDR unit tests: S parameter, single/multi material, LDR→CM convergence check
• TestLDRPolarizability test class in test_dda_solver.py
• test_auto_voxel_size_ultra precision test

Changed

• Precision targets updated based on Yurkin & Hoekstra (2007): low (0.63), medium (0.5), high (0.3), ultra (0.15)
• _prepare_dda no longer returns positions (5-tuple); positions extracted per-solve in _solve_single_wl
• Removed _apply_radiative_correction and _extract_dipoles functions (superseded by _compute_polarizability)
• auto_voxel_size docstring updated to document "ultra" precision
• NumPy dependency upper bound <2.0 removed

Fixed

• NumPy 2.x compatibility: _trapz compat shim (hasattr-based) replaces bare np.trapz/np.trapezoid in mie_solver.py and test files
• test_datastructs.py precision tests updated for new target values
• Stale test_extract_dipoles.py that referenced deleted functions rewritten

Documentation

• LDR Polarizability section with Draine & Goodman 1993 formula and 4-tier precision level table added to optical computation guide
• Mie vs DDA validation tutorial rewritten for multi-precision LDR convergence comparison
• precision parameter added to quickstart key parameters
• README updated with LDR polarizability, precision levels in features/API/example table
• validate_mie_vs_dda.py example rewritten for multi-precision convergence analysis

Tests

• All Julia-dependent tests marked @pytest.mark.slow — fast suite (-m "not slow") runs in ~1s
• Orientational averaging tests: n_dirs reduced to 2 for CI speed