From Eigenmodes to Proofs: Integrating Graph Spectral Operators with Symbolic Interpretable Reasoning