[UR][CUDA][Coverity] Don't check nvmlShutdown return in destructor (#19512)

`UR_CHECK_ERROR` throws an exception so we can't use it from a
destructor.

Since devices only get released at application shutdown, there's not
much we can do if `nvmlShutdown` reports an error, and the OS will
promptly cleanup anything left as it tears down the process anyway.

So just skip checking the error code from `nvmlShutdown`, this is
already what we do for other cleanup/shutdown in this destructor.

Fixes: #19500

Author: npmiller

unified-runtime/source/adapters/cuda/device.hpp

@@ -113,7 +113,7 @@ struct ur_device_handle_t_ : ur::cuda::handle_base {
       umfMemoryProviderDestroy(MemoryProviderShared);
     }
     if (NVMLDevice.has_value()) {
-      UR_CHECK_ERROR(nvmlShutdown());
+      nvmlShutdown();
     }
     cuDevicePrimaryCtxRelease(CuDevice);
   }