test: benchmark CI from laurentsenta

[SgtPooki]

see https://github.com/laurentsenta/helia/blob/e2a79985b78429f1abf008373794197ccc2f0f76/.github/workflows/benchmark.yml

fixes #170

[laurentsenta]

@SgtPooki thanks for looking into this! You might have a problem with reproducibility as it is right now, when I experimented, 6 runs gave 6 different results (see the Attempts button at the top), you might need self-hosted runners. cc @galargh

[SgtPooki]

Good call.

In addition to self-hosted runners, I wonder if there are more methods we could use to ensure a more accurate representation of performance across runners. Something like running a noop test and using that as a baseline to compare the actual runs in that CI action to. We could get a relative result via a normalization function that should take the environment into consideration.

Here is a good article on this i just checked out, that isnt directly useful but talks about normalization in CI environments: https://pythonspeed.com/articles/consistent-benchmarking-in-ci/

Another good read, with DENSE solution details, can be found at https://arxiv.org/pdf/1608.04295.pdf. It covers the theory behind the system used by the Julia language for benchmarking in noisy environments