argmin benchmark

[chebbyChefNEQ]

partially addresses #762

I'm adding argmin first because it's a major hotspot when constructing index. See attached perf report.

I extracted create_file function from vector_index bench, ran it in a binary, and recorded with perf. The result shows that we spend 17% of the time in argmin function.

The total time of the program is 14s + some change. It took 1+ second alone to generate and write the data to disk before index begone. Discounting the data generation overhead, we can safely say that argmin is about 20% of the time spent when build an index.

I will follow up with AVX and Neon implementation of argmin.