ggml: backend-agnostic tensor parallelism #1243

Job	Run time
ggml-ci-nvidia-cuda	42m 40s
ggml-ci-intel-openvino-gpu-low-perf	24s
ggml-ci-linux-intel-vulkan	25m 15s
ggml-ci-nvidia-vulkan-cm	35m 15s
ggml-ci-win-intel-vulkan	13m 43s
ggml-ci-nvidia-vulkan-cm2	34m 51s
	2h 32m 8s

Provide feedback