Produce benchmark reports

[kobigurk-clabs]

No description provided.

[kobigurk-clabs]

• "just add together"
• streaming batch verification
• decompressed signatures
• NIZKs? (probably not)

[gtank]

The batch verification code in #222 runs benchmarks with the following paramters:

celo-bls-snark-rs/crates/bls-crypto/benches/batch_bls.rs

Lines 12 to 17 in 13cc566

	fn batch_bls_comparison(c: &mut Criterion) {
	let mut group = c.benchmark_group("bls");
	group.sample_size(10);
	// Generate aggregate signatures on 100-validators across 1000 blocks
	const NUM_BLOCKS: usize = 300;
	const NUM_VALIDATORS: usize = 20;

These are NOT real benchmark results, they came from my laptop while doing other things, but the general shape is clear. Aggregate screening is faster than batch verification but does not provide the same security guarantees. Batch verification remains much, much faster than individually verifying signatures.

"per epoch aggregate" adds the signatures together for one epoch
"all epoch aggregate" does a single aggregate signature against many keys and hashes
"per epoch batch" uses the new code with a random linear combination of signatures and keys against a fixed message
"per epoch individual" validates every signature individually

bls/per-epoch aggregate screening
                        time:   [976.61 ms 980.90 ms 985.14 ms]
bls/all epoch aggregate screening
                        time:   [269.52 ms 270.88 ms 272.22 ms]
bls/per-epoch batch verification
                        time:   [1.6660 s 1.6676 s 1.6695 s]
bls/per-epoch individual verification
                        time:   [19.397 s 19.417 s 19.442 s]

[gtank]

I'll close this issue now but we can post more benchmarks in the future, for further optimization work or other modes.