Instead, use a global new (size_t, align_t) operator for MacOS versions
(pre 10.14 is lacking full C++17 support in the standard library).
On all other platforms/versions we expect full C++17 support.
Restructure mysimd.hpp and add support for AVX512. Changes include
- Move mysimd.hpp to ngsimd.hpp
- Expose ngsimd.hpp to NGSolve
- New namespace ngsimd
- Second template parameter (width) for SIMD class, default to the
largest width available
- Avoid raw avx register types in the interface, use SIMD<> instead
