v1.0.0

Added

• Closed-form eigendecomposition of 3x3 real symmetric matrices in pure Lean 4
  over Float, with no Mathlib dependency
• Eigenvalues via the Habera-Zilian (2025) closed-form method
• Eigenvector construction following Eberly's (2014) r0 × r1 cross-product
  convention
• Core types: SymmMat3, Vec3, Mat3, and Decomposition (with
  eigvals and eigvecs fields)
• Numerical certificate checks at maximum justified machine precision:
  residual, orthonormality, reconstruction, determinant, and eigenvalue
  ordering
• Python parity test suite validating results against NumPy/LAPACK
• Supported on Lean toolchains v4.27.0, v4.30.0, and v4.34.0-rc2