BLS optimization

### What is wrong?

There's a need to optimize BLS verifying process. Here lists some possible actions items.

- Optimize through the math part: Vitalik has a short cut for the pairing and final exponentialization
- Profiling to identify the hotspot
- Some function rewrite options (from the offline discussion of @hwwhww and @changwu-tw )
- Cython, see #50
- gmp
- sage: only py2 support 😢
- Numba

### How can it be fixed

No files or tests are named. Start by profiling the BLS verification process to identify its hotspots, then review the proposed math, Cython, GMP, Sage, and Numba options; done would require an agreed optimization scope and evidence of improved verification performance.

Written by the indexing model from the issue text.