R: fix build on aarch64 by removing a failing test

I believe this test is currently incorrect on aarch64 and expects a
warning about loss of precision with much smaller numbers than the
platform's long doubles can handle.