refactor: move model-specific operators and fusions into backend extensions (#2253)

scripts/format-code.sh

@@ -8,16 +8,19 @@ cd "$repo_root"
 for f in src/*.cpp src/*.h src/*.hpp \
          src/conditioning/*.cpp src/conditioning/*.h src/conditioning/*.hpp \
          src/core/*.cpp src/core/*.h src/core/*.hpp \
+         src/core/cpu/*.cpp src/core/cpu/*.h \
+         src/core/cuda/*.cu src/core/cuda/*.cuh src/core/cuda/*.h \
+         src/core/cuda/sageattention/*.cuh \
          src/core/probe/*.cpp src/core/probe/*.h \
+         src/core/vulkan/*.cpp src/core/vulkan/*.h src/core/vulkan/shaders/*.comp \
          src/extensions/*.cpp src/extensions/*.h src/extensions/*.hpp \
          src/pipeline/*.cpp src/pipeline/*.h \
          src/runtime/*.cpp src/runtime/*.h src/runtime/*.hpp \
          src/model/*/*.cpp src/model/*/*.h src/model/*/*.hpp \
-         src/tokenizers/*.h src/tokenizers/*.cpp src/tokenizers/vocab/*.h src/tokenizers/vocab/*.cpp \
+         src/tokenizers/*.h src/tokenizers/*.cpp \
          src/model_io/*.h src/model_io/*.cpp examples/cli/*.cpp examples/cli/*.h examples/server/*.cpp \
          examples/common/*.hpp examples/common/*.h examples/common/*.cpp; do
   [[ -e "$f" ]] || continue
-  [[ "$f" == vocab* ]] && continue
   echo "formatting '$f'"
   # if [ "$f" != "stable-diffusion.h" ]; then
   #   clang-tidy -fix -p build_linux/ "$f"