[mlir] Make cuInit() call thread-safe.

[herhut]

Nice find!

mlir/tools/mlir-cuda-runner/mlir-cuda-runner.cpp

if (auto status = (expr)) {                                                \
       return {};                                                               \
     }                                                                          \
   } while (false)
 
 OwnedBlob compilePtxToCubin(const std::string ptx, Location loc,
                             StringRef name) {
   char jitErrorBuffer[4096] = {0};
 
-  RETURN_ON_CUDA_ERROR(cuInit(0));
+  // Initialize CUDA once in a thread-safe manner.
+  static CUresult cuInitResult = [] { return cuInit(/*flags=*/0); }();
+  RETURN_ON_CUDA_ERROR(cuInitResult);
 
   // Linking requires a device context.
   CUdevice device;
   RETURN_ON_CUDA_ERROR(cuDeviceGet(&device, 0));
   CUcontext context;
   RETURN_ON_CUDA_ERROR(cuCtxCreate(&context, 0, device));
   CUlinkState linkState;