v1.1.4

What's Changed

• Use native into_split() for 9P TCP/Unix streams to avoid tokio::io::split mutex serialization by @Barre
• Fast-path single-chunk reads in chunk_store to skip scan/spawn overhead by @Barre
• Lazily allocate Notify in 9P dispatch only when Tflush actually arrives by @Barre
• Coalesce 9P writer responses through a BufWriter and drain channel per burst by @Barre
• Switch foyer caches to S3FIFO eviction to avoid per-shard write-lock contention on cache hits by @Barre

Full Changelog: v1.1.3...v1.1.4