Fix error aspect fp64 is not supported in tests of std::complex

[SergeyKopienko]

In this PR we fix the error Required aspect fp64 is not supported on the device in some tests of std::complex.

[dmitriy-sobolev]

LGTM

[danhoeflinger]

As discussed offline, I'm not sure how this should help this error. The numeric literals should all be integers, which shouldn't cause problems being confused for doubles. We already seem to protect against testing cases with doubles with our macros, so its only floats that we cast to. These should already be implicitly converting due to constexpr constructor of std::complex, so I'd expect no change to the issue.

I see these changes as harmless, but I don't understand how they help either.

[danhoeflinger]

So, I think your new changes will avoid the problem.
Here is the comment from the definition of TestUtils::infinity_val:

    // This constant was declared to avoid the substitution and evaluation of the macro INFINITY
    // inside Kernel code. Since the implementation of INFINITY macro is not regulated, there are
    // cases where temporary expressions of type double are used to calculate the value of INFINITY.
    // This lead to an error on devices where the double type is not supported.

However, it effectively just removes any testing of this for devices not supporting double. Can we pick a different baseline check for these with floating point which doesn't use TestUtils::infinity_val`?

[akukanov]

This is clearly not a proper fix. If the test is supposed to check the use of floats on a device that does not support doubles, obviously the test itself should not use doubles either.

If INFINITY does not work for some reason, there are alternatives worth checking: https://en.cppreference.com/w/cpp/types/numeric_limits/infinity, https://en.cppreference.com/w/cpp/numeric/math/HUGE_VAL

[SergeyKopienko]

So, I think your new changes will avoid the problem. Here is the comment from the definition of TestUtils::infinity_val:

    // This constant was declared to avoid the substitution and evaluation of the macro INFINITY
    // inside Kernel code. Since the implementation of INFINITY macro is not regulated, there are
    // cases where temporary expressions of type double are used to calculate the value of INFINITY.
    // This lead to an error on devices where the double type is not supported.

However, it effectively just removes any testing of this for devices not supporting double. Can we pick a different baseline check for these with floating point which doesn't use TestUtils::infinity_val`?

All other significant changes are packed into test_edges() which also required double type support.

[MikeDvorskiy]

LGTM