* vulkan : refactor/simplify buffer handling in vk_op_* functions * Combine UMA handling into ggml_vk_tensor_subbuffer