Modernize PatForward

• replace AUTO_RETURN macro with C++14 decltype(auto) return value deduction
• replace make_... functions with class template argument deduction
• replace plain owning pointers with std::unique_ptr
• use std::invoke to generically invoke callables
• pick SIMD vector size depending on AVX512F, AVX and SSE pre-processer flags
• use std::inner_product instead of explicitly writing out an unrolled inner product
• replace custom index_sequence implementation with std::index_sequence