[Attention] FA3 decode perf improvement - single mma warp group support for head dim 128 (#16864)

Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>

Author: LucasWilkinson

cmake/external_projects/vllm_flash_attn.cmake

@@ -38,7 +38,7 @@ else()
   FetchContent_Declare(
           vllm-flash-attn
           GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
-          GIT_TAG 0a721daebe4fa7149f06ecf3d3eabeb6dcd0f1fa
+          GIT_TAG e93779c59ba4905e56e5c39dc2c1904ada71fa21
           GIT_PROGRESS TRUE
           # Don't share the vllm-flash-attn build between build types
           BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn