This PR includes several changes and build performance improvements:

• added flag to CMakeLists.txt to gate option arch=native (CC @rob-p);
• user can now specify a RAM limit to build the index;
• the build_index step is now done in parallel;
• blocks of minimizer tuples are sorted in parallel using a custom parallel_sort function, coded in parallel_sort.hpp (which might also have independent interest);
• streaming of minimizers during file parsing;
• minimizers are now computed in streaming way using the "re-scan" method which is faster (and simpler) than the "monotone-queue" method used before;
• the next() method of minimizer_enumerator is templated as it used to be before.