Corpus-Guided Contrast Sets for Morphosyntactic Feature Detection in Low-Resource English Varieties