SGLang Receives KV Cache Bug Fixes for Stable Kimi K2.6 Inference

Grateful to @CloudflareDev for upstreaming the decode KV cache and Mooncake recovery fixes to SGLang! Now you can run Kimi K2.6 with decode KV cache offload under heavy concurrency without garbled outputs, and Mooncake peers recover automatically. This collab has been a joy!