On the Sweet Spot of Contrastive Views for Knowledge-enhanced Recommendation