Vectorized minmax_element: reduce insanity of sign correction

[Alex Guteniev (AlexGuteniev)]

Avoids passing sign as runtime parameter for SSE4.2 and AVX2 vectorized minmax_element.

Originally it has been done to reduce code bloat without any performance loss. SSE4.2 and AVX2 lack some unsigned instructions, so unsigned operations has to be implemented via signed ones. We applied sign correction at runtime, and shared import lib functions for signed and unsigned types.

Now that Neon has unsigned instructions, we have different entry points for signed and unsigned types to avoid runtime branches there We use compile-time sign dispatch there.. AVX-512, if implemented, will also take advantage of having unsigned instructions and will use comile-time dispatch.

This PR unifies minmax_element sign dispatch to make it compile-time everywhere.
It also renames _Sign to _Is_signed to reduce confusing.

Achivements:

• Simplification
• Unification with _Sorted_ranges
• No performance gain or loss. Some gain could have been expected, especially for signed types, but if there is such a gain, it does not stand away enough from the noise
• Some code bloat. Not measured

[Stephan T. Lavavej (StephanTLavavej)]

Thanks, I really like this simplification! 😻

[Stephan T. Lavavej (StephanTLavavej)]

I'm mirroring this to the MSVC-internal repo. Please notify me if any further changes are pushed, otherwise no action is required.

[Stephan T. Lavavej (StephanTLavavej)]

😻 🪄 💖