Make running benchmarks in CI more consistent

[pawelchcki]

This PR was made in an effort to isolate and measure performance improvements.

Background:

Due to complex nature our current benchmark have high variability in results. This means that its hard/impossible to compare the results looking for improvements or regressions.

Description:

This PR allows running multiple benchmarks in parallel and collecting + processing the results.

Basically using fork + join pattern.

Additional improvements:

In addition this PR include a few improvements to CI like:

• postgres using RAM disk as DB store
• optimized redis
• optimized mongodb
• cached bundler install and apprisal install - this shaves off 1-4 minutes of off each build
• removed checkout build - instead every build does its own git checkout + workspace is used to transfer data downstream in builds

Thus in addition to benchmark this PR optimizes the CI runtime ~2x.

Benchmarks:

All benchmarks are optional not to use too much resources, additionally they are currently limited in how much resources they use. I.e. they are split into 5 parts each with parallelism of 1. If we've had more resources we might increase parallelism to run more jobs and make our data even more trustworthy.

[delner]

Just a minor request but could we squash some of these commits? There are a lot of them, haha.

[pawelchcki]

Can you take another look at this PR @delner?

[delner]

This is a very old PR and out of date. Will close for now.