[PROJECT] Stabilize and clarify lanes API

[jfalcou]

#1193 and some other issues showed that the naming and usage of the API for retrieving the size of a wide is a bit ajar.
There was also a POC that showed promise of better API when using a NTTP instead of fixed.

As both are tied, here is a list of actions to stabilize this:

• Remove eve::scalar_cardinal as it is nto used anywhere else
• Rewrite eve::simd_value to not require eve::scalar_cardinal ([PROJECT] Cleanup basic concepts & tests #1144).
• Rename T::size() to T::lanes() so it is consistent with the eve::lane<N> inline variable
• Remove eve::cardinal_* and use eve::wide<T,N>::size() everywhere it should have been
• Rename expected/fundamental_cardinal to something better
• Implement eve::lane as a constexpr object with proper semantic
• Make eve::wide accept eve::laneobject as NTTP

This induces multiple API BREAKING CHANGES and will need to be advertised as such.
Moving forward this will require a release.

[jfalcou]

Sample POC for NTTP lane (https://godbolt.org/z/3n5jWKMb6)

#include <bit>
#include <cstdint>
#include <compare>

struct fixed_lane
{
  consteval fixed_lane(std::ptrdiff_t v) : value(v)
  {
    if( v!=0 && !std::has_single_bit( static_cast<std::size_t>(v)) )
      throw "[eve] Lane must be a power of 2";
  }

  consteval auto operator<=>(const fixed_lane&) const noexcept = default;
  friend consteval fixed_lane operator*(fixed_lane l, int n) noexcept { return {l.value * n}; }
  friend consteval fixed_lane operator*(int n, fixed_lane l) noexcept { return {l.value * n}; }
  friend consteval fixed_lane operator/(fixed_lane l, int n) noexcept { return {l.value / n}; }

  std::ptrdiff_t const value;
};

template<std::ptrdiff_t N> inline constexpr fixed_lane lane{N};

The consteval constructor with the test+throw means you can never have a non power of 2 lane and if you try stupid thign like lane * 3, it fails at compile time.