Text this: SERMixAR: Contrastive Multi-View Fusion for Robust Arabic Speech Emotion Recognition