Fix NVCC test failures involving long double

[wmaxey]

This patch fixes NVCC's complaints about using long double with some type checks:

//    Make sure the types are right
    static_assert ( cuda::std::is_same<decltype( 3h   ), cuda::std::chrono::hours>::value, "" );
    static_assert ( cuda::std::is_same<decltype( 3min ), cuda::std::chrono::minutes>::value, "" );
    static_assert ( cuda::std::is_same<decltype( 3s   ), cuda::std::chrono::seconds>::value, "" );
    static_assert ( cuda::std::is_same<decltype( 3ms  ), cuda::std::chrono::milliseconds>::value, "" );
    static_assert ( cuda::std::is_same<decltype( 3us  ), cuda::std::chrono::microseconds>::value, "" );
    static_assert ( cuda::std::is_same<decltype( 3ns  ), cuda::std::chrono::nanoseconds>::value, "" );

Original error:

2021-05-17 19:15:51+00:00| Error #20208-D: 'long double' is treated as 'double' in device code
2021-05-17 19:15:51+00:00| 
2021-05-17 19:15:51+00:00| Error #20208-D: 'long double' is treated as 'double' in device code
2021-05-17 19:15:51+00:00| 
2021-05-17 19:15:51+00:00| Error #20208-D: 'long double' is treated as 'double' in device code
2021-05-17 19:15:51+00:00| 
2021-05-17 19:15:51+00:00| Error #20208-D: 'long double' is treated as 'double' in device code
2021-05-17 19:15:51+00:00| 
2021-05-17 19:15:51+00:00| Error #20208-D: 'long double' is treated as 'double' in device code
2021-05-17 19:15:51+00:00| 
2021-05-17 19:15:51+00:00| Error #20208-D: 'long double' is treated as 'double' in device code
2021-05-17 19:15:51+00:00| 
2021-05-17 19:15:51+00:00| 6 errors detected in the compilation of "/sw/gpgpu/libcudacxx/.upstream-tests/test/std/utilities/time/time.duration/time.duration.literals/literals2.pass.cpp".
2021-05-17 19:15:51+00:00| # --error 0x1 --

According to the draft for chrono literals, the duration representation is unspecified, so this should be fine, and makes NVCC happy.
http://eel.is/c++draft/time.duration.literals

[griwes]

@brycelelbach this is an abi break, but probably one we don't care about. Do you agree with this assessment?

[brycelelbach]

It doesn't seem like a big deal, but it is an ABI break. We are reving the ABI in 2.0.0 for complex anyways, so we might as well hold this off until then.

So I would:

• First, disable the failing parts of the test, then integrate that to 1.6.0.
• Then, I would fix it by making these double instead of long double and target that for 2.0.0

But keep in mind user defined literals must take a long double.

[brycelelbach]

This looks good.

[brycelelbach]

Actually, maybe we move the ABI break from 2.0.0 to 1.6.0.

[wmaxey]

Closing as this change is included in #172.