Build shared library with runtime ISA dispatch

[r-devulap]

Summary of changes:

1. API change: all the methods will use size_t instead of int64_t for representing array sizes. This means the argsort and argselect methods will return std::vector<size_t>
2. The AVX-512 template library is built into a shared library with runtime dispatch
3. The tests and benchmarks use the shared library and now can be built without AVX-512 specific compile flags.