Use sycl::rint rather than std::rint to resolve failed tests on N…

…vidia hardware

When compiled for CUDA, `std::rint` would incorrectly round values halfway between two integers toward 0, rather than to the nearest even number as required per array API.

`sycl::rint` avoids such issues by not relying on the current rounding mode

dpctl/tensor/libtensor/include/kernels/elementwise_functions/round.hpp

@@ -81,7 +81,7 @@ template <typename argT, typename resT> struct RoundFunctor
 private:
     template <typename T> T round_func(const T &input) const
     {
-        return std::rint(input);
+        return sycl::rint(input);
     }
 };