Benchmark normalization speed

[ChristopheDuong]

Tell us about the problem you're trying to solve

Following up on #4286, we also need a test that can run normalization on a "larger" amount of data than the current integration tests to study the scalability of the generated models (whether full refresh or incremental).

See https://airbytehq.slack.com/archives/C01MFR03D5W/p1636199069220000

Describe the solution you’d like

Generate X numbers of new rows replicated in all destinations
Measure time/amount of data processed to process the X new rows

hopefully by avoiding reprocessing Y old rows from history (that may be quite large, unnecessary and expensive to work again on)

[ChristopheDuong]

As part of normalization integration tests, we should track running time for each model/query and record these in a persistent destination somewhere.

By comparing each run tests to the moving average of a few last runs, we should be able to "statistically" detect large variations and at least observe the overall progression of normalization running times for each query.