docs: add performance optimization design document for vectorless v0.3.0

Add comprehensive design document outlining performance optimization
strategies targeting millisecond-level response times. Includes detailed
plans for cache strategy optimization with semantic similarity,
incremental indexing with subtree-level updates, parallel retrieval
optimization, and memory footprint reduction.

The document covers implementation timeline, success criteria,
dependencies, and risk mitigation strategies for achieving 90%+ cache
hit rates, sub-100ms retrieval latency, and improved memory efficiency.

Author: zTgx