Simplified implementation

[jan-wassenberg]

Hi, I thought it might be useful to have a simplified (sequential) version of the code.
Please consider the attached code a patch authored and contributed by Google.

I combined all the source files into one header, reduced it to about 1100 lines and added some optional AVX-512 SIMD for classification to see how much of a difference that makes. There are some TODO comments indicating hints for future optimizations.

Note that the code is mostly de-templatized so that symbols are more easily visible in the profiler. For actual use, the Cfg struct should be parameterized on the key type again, and also passed as a parameter to the Sorter class.

[ips4o]

Hi! Thank you for attaching the implementation.