v1.0.1 — Optimization Performance

Performance improvements

• 17x faster warm-state operations through buffered persist serialization, persist skip-reload optimization, and refined fsync policy on the reconstructible cache
• Self-cleaning storage with 24-hour spill TTL, 256MB capacity ceiling, automatic orphaned temp-file sweep (runs every session)
• Criterion persist-path regression gate prevents performance regressions on cache operations
• No breaking changes — all RACC recovery semantics are byte-identical