AMD MI355X Kimi K2.5 Inference: 7.7x Throughput, Up To 15x Interactivity in 25 Days on vLLM

vLLM PR #35850 Fixed AITER MLA Dispatch on MI355X CDNA4, Unlocking Kimi K2.5 Inference Performance at TP=8, Shipped in vLLM 0.18