Leveraging Allophony in Self-Supervised Speech Models for Atypical Pronunciation Assessment