GLV framework + precomp scripts

The final piece of the puzzle, GLV. Patent expired 2 days ago. Recommended to review after rest of algorithmic optimisations have been merged. A description of the approach taken for GLV and background knowledge can be found in this hackmd article.

3/3 of a series of PRs to scipr-lab/zexe.

Features:

Batch Affine ops
Batch and ordinary w-NAF
GLV framework, GLV precomputation script.
GLV impl for both batch affine and projective mul (only for BW6)
Batch bucket addition tree
Batched MSM (1.4-1.7x bump)
Batch subgroup verification based on addition to random buckets
2D test matrix feature-gating with axes (curve, test type)
Refactor tests to use common template
Code timing instrumentation with fine grained control of functions and callers

Combined speedup (with w-NAF) for

Batched mul (with GLV): 2.3-2.7x.
Non-batch projective mul (with GLV): 2.1x. Without GLV: 1.65-1.8x and 1.4x resp.

Speedup for subgroup verification

Over old impl 25-80x.
Over new batch mul: 10-30x.

Future PRs:

Improved MSM scaling (free scaling to 100s of threads for problem sizes > 2^25). (Need to investigate if held back by Amdahl's law in serial portion of code).
Recursive batch verification that gives ~1.5x speedup for large n.
Accel-based CUDA scalar mul + load-balancing code/framework to balance between CPU and GPU. Performance is neck to neck between weak GPU (GTX 1650 GPU) and a powerful CPU (16T Ryzen 4800H that clocks a sustained 4.2GHz). Probably better than most laptop Intels.

Future work:

Extend GLV to all applicable curves https://github.com/scipr-lab/zexe/issues/267

arkworks-rs / snark

GLV framework + precomp scripts #296