Improve realism and coverage of ASV benchmarks

[IvoDD]

Is your feature request related to a problem? Please describe. Our current benchmarks don't test realistic large scale use cases.

Describe the solution you'd like A benchmarking suite which:

• Uses prod like S3 storage (current benchmarks won't catch regressions due to e.g. more back and forth iterations which in real use cases will result in slower speed due to higher latency than local tests)
• Uses large datasets (current benchmarks won't catch regressions related to e.g. many TABLE_DATA blocks)
• Is run separately from current ASV benchmarks because it will be expensive to run on every branch. We could e.g. run it nightly, weekly or on every release.

Describe alternatives you've considered Existing ASV benchmarks - we can add larger tests there but running them for every PR is likely too expensive.

Additional context Add any other context or screenshots about the feature request here.

[poodlewars]

We don't necessarily need to do end to end benchmarks.

As discussed should also consider:

• Alerting on the nightly build, so that we realise if it fails
• Full coverage of the API
• Tests with more interesting version chains (benchmarks at the moment have just one version)
• "End to end" benchmark of a "real-life" script (data loader + reader)
• Documenting the need to check benchmarks during the release process

[IvoDD]

Tests with more interesting version chains (benchmarks at the moment have just one version)

PR adding some ASV benchmarks for version chain traversal #1508