CUDA: re-use MLA K data for V in MMA FA (#19057) #29

Job	Run time
windows-cuda (12.4)	1h 16m 31s
windows (opencl-adreno, arm64, -G "Ninja Multi-Config" -D CMAKE_TOOLCHAIN_FILE=cmake/arm64-window...	2m 59s
ubuntu-22-cpu (x64, ubuntu-22.04)	6m 5s
macOS-arm64	3m 49s
windows-cuda (13.1)	1h 22m 39s
ubuntu-22-cpu (s390x, ubuntu-24.04-s390x)	1s
ubuntu-22-vulkan	9m 1s
macOS-x64	7m 50s
windows (vulkan, x64, -DGGML_VULKAN=ON, ggml-vulkan)	5m 17s
windows-cpu (arm64)	6m 6s
windows-cpu (x64)	9m 16s
windows-sycl	12m 51s
windows-hip (radeon, gfx1151;gfx1200;gfx1201;gfx1100;gfx1101;gfx1102;gfx1030;gfx1031;gfx1032)	57m 35s
openEuler-cann (x86, 310p, Release, off)	12m 44s
ios-xcode-build	33m 53s
openEuler-cann (aarch64, 310p, Release, off)	8m 49s
openEuler-cann (aarch64, 910b, Release, on)	12m 24s
openEuler-cann (x86, 910b, Release, on)	11m 16s
release	0s
	5h 59m 6s

Provide feedback