refector: optimize the usage of tensor_types #1027

Job	Run time
macOS-latest-cmake	2m 6s
windows-latest-cmake (cuda12, -DSD_CUDA=ON -DSD_BUILD_SHARED_LIBS=ON -DCMAKE_CUDA_ARCHITECTURES=9...	35m 24s
windows-latest-cmake (vulkan, -DSD_VULKAN=ON -DSD_BUILD_SHARED_LIBS=ON)	3m 38s
windows-latest-cmake (noavx, -DGGML_NATIVE=OFF -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DSD...	1m 46s
windows-latest-cmake (avx2, -DGGML_NATIVE=OFF -DGGML_AVX2=ON -DSD_BUILD_SHARED_LIBS=ON)	3m 57s
windows-latest-cmake (avx512, -DGGML_NATIVE=OFF -DGGML_AVX512=ON -DGGML_AVX=ON -DGGML_AVX2=ON -DS...	1m 56s
ubuntu-latest-cmake	1m 57s
windows-latest-cmake (avx, -DGGML_NATIVE=OFF -DGGML_AVX=ON -DGGML_AVX2=OFF -DSD_BUILD_SHARED_LIBS...	2m 0s
release	24s
	53m 8s

Provide feedback