[CI][macOS] Pin LLVM version and install lld (#4831)

Pinned LLVM to v19; cannot do the same for LLD though.

This allows us to revert #4827.

Author: antiagainst

.github/workflows/integration-tests.yml

@@ -460,7 +460,7 @@ jobs:
       - name: Install brew dependencies
         run: |
           brew update
-          brew install ccache llvm
+          brew install ccache llvm@19 lld
       - name: Compute cache keys
         id: cache-key
         run: |

.github/workflows/integration-tests.yml.in

@@ -439,7 +439,7 @@ jobs:
       - name: Install brew dependencies
         run: |
           brew update
-          brew install ccache llvm
+          brew install ccache llvm@19 lld
 
       - *compute-cache-keys-step
       - *cache-build-dependencies-step

python/setup.py

@@ -429,14 +429,11 @@ def build_extension(self, ext):
             cmake_args += [
                 "-DCMAKE_C_COMPILER=clang",
                 "-DCMAKE_CXX_COMPILER=clang++",
+                "-DCMAKE_LINKER=lld",
+                "-DCMAKE_EXE_LINKER_FLAGS=-fuse-ld=lld",
+                "-DCMAKE_MODULE_LINKER_FLAGS=-fuse-ld=lld",
+                "-DCMAKE_SHARED_LINKER_FLAGS=-fuse-ld=lld",
             ]
-            if platform.system() != "Darwin":
-                cmake_args += [
-                    "-DCMAKE_LINKER=lld",
-                    "-DCMAKE_EXE_LINKER_FLAGS=-fuse-ld=lld",
-                    "-DCMAKE_MODULE_LINKER_FLAGS=-fuse-ld=lld",
-                    "-DCMAKE_SHARED_LINKER_FLAGS=-fuse-ld=lld",
-                ]
 
         # Note that asan doesn't work with binaries that use the GPU, so this is
         # only useful for tools like triton-opt that don't run code on the GPU.