Multi-purpose Assistant for Low-level Visual Perception, from [Q-Instruct: Improving Low-level Visual Abilities for Multi-modality Foundation Models]