Text this: Enhancing Visual-Language Prompt Tuning Through Sparse Knowledge-Guided Context Optimization