Resolves #5479.

⚙️ Optimization

• Use _Is_same_and_builtin_trivially_equality_comparable in another place. Make that part common, as it repeats.
• Adjust _Could_compare_equal_to_value_type to return true for the same types. This also handles std::byte concisely.
• Adjust casts in vectorized dispatch to use _Find_arg_cast. Which only widens or truncates integers, and uses _Bit_cast for the rest. Also makes pointers non-special.

That's it!

🏁 Benchmark

The benchmark shows that:

• The changes indeed dispatch to vectorized implementation, and
• Clang auto-vectorization either does not apply, or is less efficient than manual vectorization

Only find_and_count ought to be enough. It is sufficient to show that the dispatch works.

To see that other algorithms (search_n, replace, remove, remove_copy) also benefit from the change, use their existing benchmark, and set CXXFLAGS=/arch:AVX2 /D_USE_STD_VECTOR_ALGORITHMS=0 to imitate non-optimization on integer types, and then set CXXFLAGS=/arch:AVX2 /D_USE_STD_VECTOR_ALGORITHMS=1 to see the optimization results. The newly vectorized types should perform the same, as corresponding integer types.

⏱️ Benchmark results

It was run with set CXXFLAGS=/arch:AVX2 to compete with Clang auto-vectorization fairly: the algorithms detect AVX2 at runtime.