Context-Optimized Memory Bank — Reduce AI token usage with structured documentation and cache-aware reading strategies