precondition checks in extents constructors

[nmm0]

This adds the missing precondition checks for mdspan.extents.cons

[dalg24]

We do not want to ensure that the error message matches some regex?

[nmm0]

We can at least for CPU but it won't work with CUDA. So I just decided not to since it doesn't matter a whole lot

[dalg24]

What do you mean it won't work with CUDA?
We do check error messages from the device in Kokkos so there is some level of support.

[nmm0]

So for now the CUDA precondition violation handler prints to stdout. EXPECT_DEATH checks stderr -- we could do something like:

#if defined(MDSPAN_IMPL_HAS_CUDA) || defined(MDSPAN_IMPL_HAS_HIP) || defined(MDSPAN_IMPL_HAS_SYCL)
  EXPECT_DEATH(test_precondition_indices_not_representable(), "");
#else
  EXPECT_DEATH(test_precondition_indices_not_representable(), "extent_is_representable");
#endif

We do this elsewhere. But really the precondition handler doesn't print the best errors anyway

[nmm0]

I updated to test the message on host. On device the message is kinda useless:

void Kokkos::detail::default_precondition_violation_handler(const char*, const char*, unsigned int): Assertion `0' failed.

So I think we should avoid testing device messages for now until we update our precondition violation handler on device (likely a separate PR)

[dalg24]

Oh, I see, because we call printf(...) and assert(false) instead of __assert_fail(...) like we do in Kokkos