#407 Cache vector index in memory for brain search #409

Problem

CodeVectorIndex::load_or_create() is called every cora brain query, loading the usearch index file from disk (~100-150ms I/O).

Location

src/index/brain.rs:174-184 -- inside vector_search()
src/index/vector.rs:57-72 -- load_or_create reads entire file

Solution

Implement OnceLock<CodeVectorIndex> or lazy static to cache the vector index in memory. Load once, reuse for all subsequent queries.

Expected Impact

~80% search speedup. Target: 250ms to 50-80ms.