Add benchmark.js and github-action-benchmark

[ryanio]

This PR formats our existing benchmarks into a benchmark.js suite to ensure statistical significance of benchmark runs, and adds github-action-benchmark to compare and keep track of results.

[coveralls]

Coverage remained the same at 95.29% when pulling 3947fdf on benchmarkjs into aa9e47c on master.

[ryanio]

Sample results from run:

1k-3-32-ran x 20,597 ops/sec ±23.17% (75 runs sampled)
1k-5-32-ran x 24,254 ops/sec ±2.16% (81 runs sampled)
1k-9-32-ran x 22,278 ops/sec ±2.52% (75 runs sampled)
1k-1k-32-ran x 9,052 ops/sec ±44.07% (33 runs sampled)
1k-1k-32-mir x 22,115 ops/sec ±5.47% (67 runs sampled)
Checkpointing: 100 iterations x 1,066 ops/sec ±6.10% (69 runs sampled)
Checkpointing: 500 iterations x 107 ops/sec ±107.49% (47 runs sampled)
Checkpointing: 1000 iterations x 110 ops/sec ±8.79% (58 runs sampled)
Checkpointing: 5000 iterations x 23.07 ops/sec ±8.41% (32 runs sampled)

Some interesting results here:

• You can see how quickly the checkpointing slows down with larger iterations.
• 1k-1k-32-ran slowdown above seems out of place with 44% amplitude, might be an outlier that normalizes with future runs.
  • Prior run results on my local machine:
    • 1k-1k-32-ran x 11,990 ops/sec ±11.70% (47 runs sampled)
    • 1k-1k-32-ran x 13,377 ops/sec ±12.03% (57 runs sampled)

[s1na]

Something else that'd be interesting to see is how checkpointing performance changes with the size of data to be commited. Here that size is one value (each value is committed separately).

[s1na]

Something else that'd be interesting for finding the slowdown is a benchmark that does checkpoint, inserts a bunch of stuff and then does revert instead of commit.

Comparing this bench on v4 and v3 should tell us whether the slowdown is happening in commit (which is my suspicion) or not.

[holgerd77]

Rebased this.

[holgerd77]

LGTM

Coverage is not working right now. Will merge nevertheless since the library has to be switched to codecov anyhow to prepare for the monorepo transition.