Fix potential termination with noexcept specifier

[yosh-matsuda]

Checklist

• The title and commit message(s) are descriptive.
• Small commits made to fix your PR have been squashed to avoid history pollution.
• Tests have been added for new features or bug fixes.
• API of new functions and classes are documented.

Description

I found xt::broadcast_error is not caught and std::terminate is called for assign operators with xt::xarray_adaptor.

To reproduce:

void test()
{
    using namespace xt;
    using vec_type = std::vector<int>;
    using xarray_adaptor_type = xarray_adaptor<vec_type, layout_type::dynamic>;

    auto v1 = vec_type{1, 2, 3, 4};
    auto xa1 = xarray_adaptor_type(v1, {4});

    auto v2 = vec_type{10, 20, 30, 40};
    auto xa2 = xarray_adaptor_type(v2, {2, 2});

    try
    {
        // expect broadcast error
        xa1 += xa2;
    }
    catch (const xt::broadcast_error& e)
    {
        // xt::broadcast_error will NOT be caught
        // std::terminate called from compiler-generated code
        std::println("broadcast error: {}\n", e.what());
    }
}

It seems that this is because xt::broadcast_error exception is thrown within noexcept specified function.
Here:

    template <class D>
    inline auto xconst_accessible<D>::size() const noexcept -> size_type
    {
        return compute_size(derived_cast().shape());
    }

    inline void xfunction<F, CT...>::compute_cached_shape() const
    {
        static_assert(!detail::is_fixed<shape_type>::value, "Calling compute_cached_shape on fixed!");

        m_cache.shape = uninitialized_shape<xindex_type_t<inner_shape_type>>(compute_dimension());
        m_cache.is_trivial = broadcast_shape(m_cache.shape, false);
        m_cache.is_initialized = true;
    }

    template <class F, class... CT>
    inline auto xfunction<F, CT...>::shape() const -> const inner_shape_type& 
    {
        if constexpr (!detail::is_fixed<inner_shape_type>::value)
        {
            if (!m_cache.is_initialized)
            {
                compute_cached_shape();
            }
        }
        return m_cache.shape;
    }

derived_cast().shape() possibly throw exception and noexcept should be replaced by noexcept(noexcept(...)) to fix it.

[JohanMabille]

Thanks! The error in te CI is due to a well-known concurrency issue.